diff mbox series

net/virtio: add platform memory ordering feature support

Message ID 20181214153812.3878-1-i.maximets@samsung.com (mailing list archive)
State Superseded, archived
Delegated to: Maxime Coquelin
Headers
Series net/virtio: add platform memory ordering feature support |

Checks

Context Check Description
ci/checkpatch success coding style OK
ci/Intel-compilation success Compilation OK
ci/mellanox-Performance-Testing success Performance Testing PASS
ci/intel-Performance-Testing success Performance Testing PASS

Commit Message

Ilya Maximets Dec. 14, 2018, 3:38 p.m. UTC 
  VIRTIO_F_ORDER_PLATFORM is required to use proper memory barriers
in case of HW vhost implementations like vDPA.

DMA barriers (rte_cio_*) are sufficent for that purpose.

Previously known as VIRTIO_F_IO_BARRIER.

Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
---

Based on "[RFC] net/virtio: use real barriers for vDPA".

RFC --> PATCH:
  * Dropped vendor-specific hack to determine if we need real barriers.
  * Added VIRTIO_F_ORDER_PLATFORM feature definition and checking.

Note: Patch to change the name of the feature from VIRTIO_F_IO_BARRIER
      to VIRTIO_F_ORDER_PLATFORM is not merged yet:
      https://www.mail-archive.com/virtio-dev@lists.oasis-open.org/msg04114.html

 drivers/net/virtio/virtio_ethdev.c |  2 ++
 drivers/net/virtio/virtio_ethdev.h |  3 ++-
 drivers/net/virtio/virtio_pci.h    |  7 ++++++
 drivers/net/virtio/virtio_rxtx.c   | 14 ++++++------
 drivers/net/virtio/virtqueue.h     | 35 +++++++++++++++++++++++++-----
 5 files changed, 48 insertions(+), 13 deletions(-)

Comments

Michael S. Tsirkin Dec. 14, 2018, 5 p.m. UTC | #1 
On Fri, Dec 14, 2018 at 06:38:12PM +0300, Ilya Maximets wrote:
> VIRTIO_F_ORDER_PLATFORM is required to use proper memory barriers
> in case of HW vhost implementations like vDPA.
> 
> DMA barriers (rte_cio_*) are sufficent for that purpose.
> 
> Previously known as VIRTIO_F_IO_BARRIER.
> 
> Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
> ---
> 
> Based on "[RFC] net/virtio: use real barriers for vDPA".
> 
> RFC --> PATCH:
>   * Dropped vendor-specific hack to determine if we need real barriers.
>   * Added VIRTIO_F_ORDER_PLATFORM feature definition and checking.
> 
> Note: Patch to change the name of the feature from VIRTIO_F_IO_BARRIER
>       to VIRTIO_F_ORDER_PLATFORM is not merged yet:
>       https://www.mail-archive.com/virtio-dev@lists.oasis-open.org/msg04114.html
> 
>  drivers/net/virtio/virtio_ethdev.c |  2 ++
>  drivers/net/virtio/virtio_ethdev.h |  3 ++-
>  drivers/net/virtio/virtio_pci.h    |  7 ++++++
>  drivers/net/virtio/virtio_rxtx.c   | 14 ++++++------
>  drivers/net/virtio/virtqueue.h     | 35 +++++++++++++++++++++++++-----
>  5 files changed, 48 insertions(+), 13 deletions(-)
> 
> diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
> index cb2b2e0bf..5ae7a9650 100644
> --- a/drivers/net/virtio/virtio_ethdev.c
> +++ b/drivers/net/virtio/virtio_ethdev.c
> @@ -1474,6 +1474,8 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
>  	if (virtio_negotiate_features(hw, req_features) < 0)
>  		return -1;
>  
> +	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
> +
>  	if (!hw->virtio_user_dev) {
>  		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
>  		rte_eth_copy_pci_info(eth_dev, pci_dev);
> diff --git a/drivers/net/virtio/virtio_ethdev.h b/drivers/net/virtio/virtio_ethdev.h
> index e0f80e5a4..c098e5ac0 100644
> --- a/drivers/net/virtio/virtio_ethdev.h
> +++ b/drivers/net/virtio/virtio_ethdev.h
> @@ -34,7 +34,8 @@
>  	 1u << VIRTIO_RING_F_INDIRECT_DESC |    \
>  	 1ULL << VIRTIO_F_VERSION_1       |	\
>  	 1ULL << VIRTIO_F_IN_ORDER        |	\
> -	 1ULL << VIRTIO_F_IOMMU_PLATFORM)
> +	 1ULL << VIRTIO_F_IOMMU_PLATFORM  |	\
> +	 1ULL << VIRTIO_F_ORDER_PLATFORM)
>  
>  #define VIRTIO_PMD_SUPPORTED_GUEST_FEATURES	\
>  	(VIRTIO_PMD_DEFAULT_GUEST_FEATURES |	\
> diff --git a/drivers/net/virtio/virtio_pci.h b/drivers/net/virtio/virtio_pci.h
> index e961a58ca..e2f096185 100644
> --- a/drivers/net/virtio/virtio_pci.h
> +++ b/drivers/net/virtio/virtio_pci.h
> @@ -128,6 +128,12 @@ struct virtnet_ctl;
>   */
>  #define VIRTIO_F_IN_ORDER 35
>  
> +/*
> + * This feature indicates that memory accesses by the driver and the device
> + * are ordered in a way described by the platform.
> + */
> +#define VIRTIO_F_ORDER_PLATFORM 36
> +
>  /* The Guest publishes the used index for which it expects an interrupt
>   * at the end of the avail ring. Host should ignore the avail->flags field. */
>  /* The Host publishes the avail index for which it expects a kick
> @@ -240,6 +246,7 @@ struct virtio_hw {
>  	uint8_t     use_simple_rx;
>  	uint8_t     use_inorder_rx;
>  	uint8_t     use_inorder_tx;
> +	uint8_t     weak_barriers;
>  	bool        has_tx_offload;
>  	bool        has_rx_offload;
>  	uint16_t    port_id;
> diff --git a/drivers/net/virtio/virtio_rxtx.c b/drivers/net/virtio/virtio_rxtx.c
> index cb8f89f18..66195bf47 100644
> --- a/drivers/net/virtio/virtio_rxtx.c
> +++ b/drivers/net/virtio/virtio_rxtx.c
> @@ -906,7 +906,7 @@ virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
>  
>  	nb_used = VIRTQUEUE_NUSED(vq);
>  
> -	virtio_rmb();
> +	virtio_rmb(hw->weak_barriers);
>  
>  	num = likely(nb_used <= nb_pkts) ? nb_used : nb_pkts;
>  	if (unlikely(num > VIRTIO_MBUF_BURST_SZ))
> @@ -1017,7 +1017,7 @@ virtio_recv_mergeable_pkts_inorder(void *rx_queue,
>  	nb_used = RTE_MIN(nb_used, nb_pkts);
>  	nb_used = RTE_MIN(nb_used, VIRTIO_MBUF_BURST_SZ);
>  
> -	virtio_rmb();
> +	virtio_rmb(hw->weak_barriers);
>  
>  	PMD_RX_LOG(DEBUG, "used:%d", nb_used);
>  
> @@ -1202,7 +1202,7 @@ virtio_recv_mergeable_pkts(void *rx_queue,
>  
>  	nb_used = VIRTQUEUE_NUSED(vq);
>  
> -	virtio_rmb();
> +	virtio_rmb(hw->weak_barriers);
>  
>  	PMD_RX_LOG(DEBUG, "used:%d", nb_used);
>  
> @@ -1365,7 +1365,7 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
>  	PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
>  	nb_used = VIRTQUEUE_NUSED(vq);
>  
> -	virtio_rmb();
> +	virtio_rmb(hw->weak_barriers);
>  	if (likely(nb_used > vq->vq_nentries - vq->vq_free_thresh))
>  		virtio_xmit_cleanup(vq, nb_used);
>  
> @@ -1407,7 +1407,7 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
>  		/* Positive value indicates it need free vring descriptors */
>  		if (unlikely(need > 0)) {
>  			nb_used = VIRTQUEUE_NUSED(vq);
> -			virtio_rmb();
> +			virtio_rmb(hw->weak_barriers);
>  			need = RTE_MIN(need, (int)nb_used);
>  
>  			virtio_xmit_cleanup(vq, need);
> @@ -1463,7 +1463,7 @@ virtio_xmit_pkts_inorder(void *tx_queue,
>  	PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
>  	nb_used = VIRTQUEUE_NUSED(vq);
>  
> -	virtio_rmb();
> +	virtio_rmb(hw->weak_barriers);
>  	if (likely(nb_used > vq->vq_nentries - vq->vq_free_thresh))
>  		virtio_xmit_cleanup_inorder(vq, nb_used);
>  
> @@ -1511,7 +1511,7 @@ virtio_xmit_pkts_inorder(void *tx_queue,
>  		need = slots - vq->vq_free_cnt;
>  		if (unlikely(need > 0)) {
>  			nb_used = VIRTQUEUE_NUSED(vq);
> -			virtio_rmb();
> +			virtio_rmb(hw->weak_barriers);
>  			need = RTE_MIN(need, (int)nb_used);
>  
>  			virtio_xmit_cleanup_inorder(vq, need);
> diff --git a/drivers/net/virtio/virtqueue.h b/drivers/net/virtio/virtqueue.h
> index 26518ed98..6b9055a1f 100644
> --- a/drivers/net/virtio/virtqueue.h
> +++ b/drivers/net/virtio/virtqueue.h
> @@ -19,15 +19,40 @@
>  struct rte_mbuf;
>  
>  /*
> - * Per virtio_config.h in Linux.
> + * Per virtio_ring.h in Linux.
>   *     For virtio_pci on SMP, we don't need to order with respect to MMIO
>   *     accesses through relaxed memory I/O windows, so smp_mb() et al are
>   *     sufficient.
>   *
> + *     For using virtio to talk to real devices (eg. vDPA) we do need real
> + *     barriers.
>   */
> -#define virtio_mb()	rte_smp_mb()
> -#define virtio_rmb()	rte_smp_rmb()
> -#define virtio_wmb()	rte_smp_wmb()
> +static inline void
> +virtio_mb(uint8_t weak_barriers)
> +{
> +	if (weak_barriers)
> +		rte_smp_mb();
> +	else
> +		rte_mb();
> +}
> +

Why doesn't rte_cio_rmb exit?

> +static inline void
> +virtio_rmb(uint8_t weak_barriers)
> +{
> +	if (weak_barriers)
> +		rte_smp_rmb();
> +	else
> +		rte_cio_rmb();
> +}
> +
> +static inline void
> +virtio_wmb(uint8_t weak_barriers)
> +{
> +	if (weak_barriers)
> +		rte_smp_wmb();
> +	else
> +		rte_cio_wmb();
> +}
>  
>  #ifdef RTE_PMD_PACKET_PREFETCH
>  #define rte_packet_prefetch(p)  rte_prefetch1(p)
> @@ -312,7 +337,7 @@ void vq_ring_free_inorder(struct virtqueue *vq, uint16_t desc_idx,
>  static inline void
>  vq_update_avail_idx(struct virtqueue *vq)
>  {
> -	virtio_wmb();
> +	virtio_wmb(vq->hw->weak_barriers);
>  	vq->vq_ring.avail->idx = vq->vq_avail_idx;
>  }
>  
> -- 
> 2.17.1
Ilya Maximets Dec. 14, 2018, 5:23 p.m. UTC | #2 
On 14.12.2018 20:00, Michael S. Tsirkin wrote:
> On Fri, Dec 14, 2018 at 06:38:12PM +0300, Ilya Maximets wrote:
>> VIRTIO_F_ORDER_PLATFORM is required to use proper memory barriers
>> in case of HW vhost implementations like vDPA.
>>
>> DMA barriers (rte_cio_*) are sufficent for that purpose.
>>
>> Previously known as VIRTIO_F_IO_BARRIER.
>>
>> Signed-off-by: Ilya Maximets <i.maximets@samsung.com>
>> ---
>>
>> Based on "[RFC] net/virtio: use real barriers for vDPA".
>>
>> RFC --> PATCH:
>>   * Dropped vendor-specific hack to determine if we need real barriers.
>>   * Added VIRTIO_F_ORDER_PLATFORM feature definition and checking.
>>
>> Note: Patch to change the name of the feature from VIRTIO_F_IO_BARRIER
>>       to VIRTIO_F_ORDER_PLATFORM is not merged yet:
>>       https://www.mail-archive.com/virtio-dev@lists.oasis-open.org/msg04114.html
>>
>>  drivers/net/virtio/virtio_ethdev.c |  2 ++
>>  drivers/net/virtio/virtio_ethdev.h |  3 ++-
>>  drivers/net/virtio/virtio_pci.h    |  7 ++++++
>>  drivers/net/virtio/virtio_rxtx.c   | 14 ++++++------
>>  drivers/net/virtio/virtqueue.h     | 35 +++++++++++++++++++++++++-----
>>  5 files changed, 48 insertions(+), 13 deletions(-)
>>
>> diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
>> index cb2b2e0bf..5ae7a9650 100644
>> --- a/drivers/net/virtio/virtio_ethdev.c
>> +++ b/drivers/net/virtio/virtio_ethdev.c
>> @@ -1474,6 +1474,8 @@ virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
>>  	if (virtio_negotiate_features(hw, req_features) < 0)
>>  		return -1;
>>  
>> +	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
>> +
>>  	if (!hw->virtio_user_dev) {
>>  		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
>>  		rte_eth_copy_pci_info(eth_dev, pci_dev);
>> diff --git a/drivers/net/virtio/virtio_ethdev.h b/drivers/net/virtio/virtio_ethdev.h
>> index e0f80e5a4..c098e5ac0 100644
>> --- a/drivers/net/virtio/virtio_ethdev.h
>> +++ b/drivers/net/virtio/virtio_ethdev.h
>> @@ -34,7 +34,8 @@
>>  	 1u << VIRTIO_RING_F_INDIRECT_DESC |    \
>>  	 1ULL << VIRTIO_F_VERSION_1       |	\
>>  	 1ULL << VIRTIO_F_IN_ORDER        |	\
>> -	 1ULL << VIRTIO_F_IOMMU_PLATFORM)
>> +	 1ULL << VIRTIO_F_IOMMU_PLATFORM  |	\
>> +	 1ULL << VIRTIO_F_ORDER_PLATFORM)
>>  
>>  #define VIRTIO_PMD_SUPPORTED_GUEST_FEATURES	\
>>  	(VIRTIO_PMD_DEFAULT_GUEST_FEATURES |	\
>> diff --git a/drivers/net/virtio/virtio_pci.h b/drivers/net/virtio/virtio_pci.h
>> index e961a58ca..e2f096185 100644
>> --- a/drivers/net/virtio/virtio_pci.h
>> +++ b/drivers/net/virtio/virtio_pci.h
>> @@ -128,6 +128,12 @@ struct virtnet_ctl;
>>   */
>>  #define VIRTIO_F_IN_ORDER 35
>>  
>> +/*
>> + * This feature indicates that memory accesses by the driver and the device
>> + * are ordered in a way described by the platform.
>> + */
>> +#define VIRTIO_F_ORDER_PLATFORM 36
>> +
>>  /* The Guest publishes the used index for which it expects an interrupt
>>   * at the end of the avail ring. Host should ignore the avail->flags field. */
>>  /* The Host publishes the avail index for which it expects a kick
>> @@ -240,6 +246,7 @@ struct virtio_hw {
>>  	uint8_t     use_simple_rx;
>>  	uint8_t     use_inorder_rx;
>>  	uint8_t     use_inorder_tx;
>> +	uint8_t     weak_barriers;
>>  	bool        has_tx_offload;
>>  	bool        has_rx_offload;
>>  	uint16_t    port_id;
>> diff --git a/drivers/net/virtio/virtio_rxtx.c b/drivers/net/virtio/virtio_rxtx.c
>> index cb8f89f18..66195bf47 100644
>> --- a/drivers/net/virtio/virtio_rxtx.c
>> +++ b/drivers/net/virtio/virtio_rxtx.c
>> @@ -906,7 +906,7 @@ virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
>>  
>>  	nb_used = VIRTQUEUE_NUSED(vq);
>>  
>> -	virtio_rmb();
>> +	virtio_rmb(hw->weak_barriers);
>>  
>>  	num = likely(nb_used <= nb_pkts) ? nb_used : nb_pkts;
>>  	if (unlikely(num > VIRTIO_MBUF_BURST_SZ))
>> @@ -1017,7 +1017,7 @@ virtio_recv_mergeable_pkts_inorder(void *rx_queue,
>>  	nb_used = RTE_MIN(nb_used, nb_pkts);
>>  	nb_used = RTE_MIN(nb_used, VIRTIO_MBUF_BURST_SZ);
>>  
>> -	virtio_rmb();
>> +	virtio_rmb(hw->weak_barriers);
>>  
>>  	PMD_RX_LOG(DEBUG, "used:%d", nb_used);
>>  
>> @@ -1202,7 +1202,7 @@ virtio_recv_mergeable_pkts(void *rx_queue,
>>  
>>  	nb_used = VIRTQUEUE_NUSED(vq);
>>  
>> -	virtio_rmb();
>> +	virtio_rmb(hw->weak_barriers);
>>  
>>  	PMD_RX_LOG(DEBUG, "used:%d", nb_used);
>>  
>> @@ -1365,7 +1365,7 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
>>  	PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
>>  	nb_used = VIRTQUEUE_NUSED(vq);
>>  
>> -	virtio_rmb();
>> +	virtio_rmb(hw->weak_barriers);
>>  	if (likely(nb_used > vq->vq_nentries - vq->vq_free_thresh))
>>  		virtio_xmit_cleanup(vq, nb_used);
>>  
>> @@ -1407,7 +1407,7 @@ virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
>>  		/* Positive value indicates it need free vring descriptors */
>>  		if (unlikely(need > 0)) {
>>  			nb_used = VIRTQUEUE_NUSED(vq);
>> -			virtio_rmb();
>> +			virtio_rmb(hw->weak_barriers);
>>  			need = RTE_MIN(need, (int)nb_used);
>>  
>>  			virtio_xmit_cleanup(vq, need);
>> @@ -1463,7 +1463,7 @@ virtio_xmit_pkts_inorder(void *tx_queue,
>>  	PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
>>  	nb_used = VIRTQUEUE_NUSED(vq);
>>  
>> -	virtio_rmb();
>> +	virtio_rmb(hw->weak_barriers);
>>  	if (likely(nb_used > vq->vq_nentries - vq->vq_free_thresh))
>>  		virtio_xmit_cleanup_inorder(vq, nb_used);
>>  
>> @@ -1511,7 +1511,7 @@ virtio_xmit_pkts_inorder(void *tx_queue,
>>  		need = slots - vq->vq_free_cnt;
>>  		if (unlikely(need > 0)) {
>>  			nb_used = VIRTQUEUE_NUSED(vq);
>> -			virtio_rmb();
>> +			virtio_rmb(hw->weak_barriers);
>>  			need = RTE_MIN(need, (int)nb_used);
>>  
>>  			virtio_xmit_cleanup_inorder(vq, need);
>> diff --git a/drivers/net/virtio/virtqueue.h b/drivers/net/virtio/virtqueue.h
>> index 26518ed98..6b9055a1f 100644
>> --- a/drivers/net/virtio/virtqueue.h
>> +++ b/drivers/net/virtio/virtqueue.h
>> @@ -19,15 +19,40 @@
>>  struct rte_mbuf;
>>  
>>  /*
>> - * Per virtio_config.h in Linux.
>> + * Per virtio_ring.h in Linux.
>>   *     For virtio_pci on SMP, we don't need to order with respect to MMIO
>>   *     accesses through relaxed memory I/O windows, so smp_mb() et al are
>>   *     sufficient.
>>   *
>> + *     For using virtio to talk to real devices (eg. vDPA) we do need real
>> + *     barriers.
>>   */
>> -#define virtio_mb()	rte_smp_mb()
>> -#define virtio_rmb()	rte_smp_rmb()
>> -#define virtio_wmb()	rte_smp_wmb()
>> +static inline void
>> +virtio_mb(uint8_t weak_barriers)
>> +{
>> +	if (weak_barriers)
>> +		rte_smp_mb();
>> +	else
>> +		rte_mb();
>> +}
>> +
> 
> Why doesn't rte_cio_rmb exit?

Assuming your question was "Why doesn't rte_cio_mb exist?".
I guess 'cio' barriers was copied from 'dma_' barriers from kernel.
And 'rte_cio_mb' does not exist because there is no 'dma_mb' in kernel.

OTOH, maybe we can use 'rte_io_mb' here to be more consistent, but it
equals to the 'rte_mb' on all supported architectures. So, I'm not sure
if it's needed.

> 
>> +static inline void
>> +virtio_rmb(uint8_t weak_barriers)
>> +{
>> +	if (weak_barriers)
>> +		rte_smp_rmb();
>> +	else
>> +		rte_cio_rmb();
>> +}
>> +
>> +static inline void
>> +virtio_wmb(uint8_t weak_barriers)
>> +{
>> +	if (weak_barriers)
>> +		rte_smp_wmb();
>> +	else
>> +		rte_cio_wmb();
>> +}
>>  
>>  #ifdef RTE_PMD_PACKET_PREFETCH
>>  #define rte_packet_prefetch(p)  rte_prefetch1(p)
>> @@ -312,7 +337,7 @@ void vq_ring_free_inorder(struct virtqueue *vq, uint16_t desc_idx,
>>  static inline void
>>  vq_update_avail_idx(struct virtqueue *vq)
>>  {
>> -	virtio_wmb();
>> +	virtio_wmb(vq->hw->weak_barriers);
>>  	vq->vq_ring.avail->idx = vq->vq_avail_idx;
>>  }
>>  
>> -- 
>> 2.17.1
> 
>
diff mbox series

Patch

diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
index cb2b2e0bf..5ae7a9650 100644
--- a/drivers/net/virtio/virtio_ethdev.c
+++ b/drivers/net/virtio/virtio_ethdev.c
@@ -1474,6 +1474,8 @@  virtio_init_device(struct rte_eth_dev *eth_dev, uint64_t req_features)
 	if (virtio_negotiate_features(hw, req_features) < 0)
 		return -1;
 
+	hw->weak_barriers = !vtpci_with_feature(hw, VIRTIO_F_ORDER_PLATFORM);
+
 	if (!hw->virtio_user_dev) {
 		pci_dev = RTE_ETH_DEV_TO_PCI(eth_dev);
 		rte_eth_copy_pci_info(eth_dev, pci_dev);
diff --git a/drivers/net/virtio/virtio_ethdev.h b/drivers/net/virtio/virtio_ethdev.h
index e0f80e5a4..c098e5ac0 100644
--- a/drivers/net/virtio/virtio_ethdev.h
+++ b/drivers/net/virtio/virtio_ethdev.h
@@ -34,7 +34,8 @@ 
 	 1u << VIRTIO_RING_F_INDIRECT_DESC |    \
 	 1ULL << VIRTIO_F_VERSION_1       |	\
 	 1ULL << VIRTIO_F_IN_ORDER        |	\
-	 1ULL << VIRTIO_F_IOMMU_PLATFORM)
+	 1ULL << VIRTIO_F_IOMMU_PLATFORM  |	\
+	 1ULL << VIRTIO_F_ORDER_PLATFORM)
 
 #define VIRTIO_PMD_SUPPORTED_GUEST_FEATURES	\
 	(VIRTIO_PMD_DEFAULT_GUEST_FEATURES |	\
diff --git a/drivers/net/virtio/virtio_pci.h b/drivers/net/virtio/virtio_pci.h
index e961a58ca..e2f096185 100644
--- a/drivers/net/virtio/virtio_pci.h
+++ b/drivers/net/virtio/virtio_pci.h
@@ -128,6 +128,12 @@  struct virtnet_ctl;
  */
 #define VIRTIO_F_IN_ORDER 35
 
+/*
+ * This feature indicates that memory accesses by the driver and the device
+ * are ordered in a way described by the platform.
+ */
+#define VIRTIO_F_ORDER_PLATFORM 36
+
 /* The Guest publishes the used index for which it expects an interrupt
  * at the end of the avail ring. Host should ignore the avail->flags field. */
 /* The Host publishes the avail index for which it expects a kick
@@ -240,6 +246,7 @@  struct virtio_hw {
 	uint8_t     use_simple_rx;
 	uint8_t     use_inorder_rx;
 	uint8_t     use_inorder_tx;
+	uint8_t     weak_barriers;
 	bool        has_tx_offload;
 	bool        has_rx_offload;
 	uint16_t    port_id;
diff --git a/drivers/net/virtio/virtio_rxtx.c b/drivers/net/virtio/virtio_rxtx.c
index cb8f89f18..66195bf47 100644
--- a/drivers/net/virtio/virtio_rxtx.c
+++ b/drivers/net/virtio/virtio_rxtx.c
@@ -906,7 +906,7 @@  virtio_recv_pkts(void *rx_queue, struct rte_mbuf **rx_pkts, uint16_t nb_pkts)
 
 	nb_used = VIRTQUEUE_NUSED(vq);
 
-	virtio_rmb();
+	virtio_rmb(hw->weak_barriers);
 
 	num = likely(nb_used <= nb_pkts) ? nb_used : nb_pkts;
 	if (unlikely(num > VIRTIO_MBUF_BURST_SZ))
@@ -1017,7 +1017,7 @@  virtio_recv_mergeable_pkts_inorder(void *rx_queue,
 	nb_used = RTE_MIN(nb_used, nb_pkts);
 	nb_used = RTE_MIN(nb_used, VIRTIO_MBUF_BURST_SZ);
 
-	virtio_rmb();
+	virtio_rmb(hw->weak_barriers);
 
 	PMD_RX_LOG(DEBUG, "used:%d", nb_used);
 
@@ -1202,7 +1202,7 @@  virtio_recv_mergeable_pkts(void *rx_queue,
 
 	nb_used = VIRTQUEUE_NUSED(vq);
 
-	virtio_rmb();
+	virtio_rmb(hw->weak_barriers);
 
 	PMD_RX_LOG(DEBUG, "used:%d", nb_used);
 
@@ -1365,7 +1365,7 @@  virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 	PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
 	nb_used = VIRTQUEUE_NUSED(vq);
 
-	virtio_rmb();
+	virtio_rmb(hw->weak_barriers);
 	if (likely(nb_used > vq->vq_nentries - vq->vq_free_thresh))
 		virtio_xmit_cleanup(vq, nb_used);
 
@@ -1407,7 +1407,7 @@  virtio_xmit_pkts(void *tx_queue, struct rte_mbuf **tx_pkts, uint16_t nb_pkts)
 		/* Positive value indicates it need free vring descriptors */
 		if (unlikely(need > 0)) {
 			nb_used = VIRTQUEUE_NUSED(vq);
-			virtio_rmb();
+			virtio_rmb(hw->weak_barriers);
 			need = RTE_MIN(need, (int)nb_used);
 
 			virtio_xmit_cleanup(vq, need);
@@ -1463,7 +1463,7 @@  virtio_xmit_pkts_inorder(void *tx_queue,
 	PMD_TX_LOG(DEBUG, "%d packets to xmit", nb_pkts);
 	nb_used = VIRTQUEUE_NUSED(vq);
 
-	virtio_rmb();
+	virtio_rmb(hw->weak_barriers);
 	if (likely(nb_used > vq->vq_nentries - vq->vq_free_thresh))
 		virtio_xmit_cleanup_inorder(vq, nb_used);
 
@@ -1511,7 +1511,7 @@  virtio_xmit_pkts_inorder(void *tx_queue,
 		need = slots - vq->vq_free_cnt;
 		if (unlikely(need > 0)) {
 			nb_used = VIRTQUEUE_NUSED(vq);
-			virtio_rmb();
+			virtio_rmb(hw->weak_barriers);
 			need = RTE_MIN(need, (int)nb_used);
 
 			virtio_xmit_cleanup_inorder(vq, need);
diff --git a/drivers/net/virtio/virtqueue.h b/drivers/net/virtio/virtqueue.h
index 26518ed98..6b9055a1f 100644
--- a/drivers/net/virtio/virtqueue.h
+++ b/drivers/net/virtio/virtqueue.h
@@ -19,15 +19,40 @@ 
 struct rte_mbuf;
 
 /*
- * Per virtio_config.h in Linux.
+ * Per virtio_ring.h in Linux.
  *     For virtio_pci on SMP, we don't need to order with respect to MMIO
  *     accesses through relaxed memory I/O windows, so smp_mb() et al are
  *     sufficient.
  *
+ *     For using virtio to talk to real devices (eg. vDPA) we do need real
+ *     barriers.
  */
-#define virtio_mb()	rte_smp_mb()
-#define virtio_rmb()	rte_smp_rmb()
-#define virtio_wmb()	rte_smp_wmb()
+static inline void
+virtio_mb(uint8_t weak_barriers)
+{
+	if (weak_barriers)
+		rte_smp_mb();
+	else
+		rte_mb();
+}
+
+static inline void
+virtio_rmb(uint8_t weak_barriers)
+{
+	if (weak_barriers)
+		rte_smp_rmb();
+	else
+		rte_cio_rmb();
+}
+
+static inline void
+virtio_wmb(uint8_t weak_barriers)
+{
+	if (weak_barriers)
+		rte_smp_wmb();
+	else
+		rte_cio_wmb();
+}
 
 #ifdef RTE_PMD_PACKET_PREFETCH
 #define rte_packet_prefetch(p)  rte_prefetch1(p)
@@ -312,7 +337,7 @@  void vq_ring_free_inorder(struct virtqueue *vq, uint16_t desc_idx,
 static inline void
 vq_update_avail_idx(struct virtqueue *vq)
 {
-	virtio_wmb();
+	virtio_wmb(vq->hw->weak_barriers);
 	vq->vq_ring.avail->idx = vq->vq_avail_idx;
 }