[v4] net/virtio: set offload flag for jumbo frames

Message ID 20190201100348.6730-1-jfreimann@redhat.com
State Superseded, archived
Delegated to: Maxime Coquelin
Headers show
Series
  • [v4] net/virtio: set offload flag for jumbo frames
Related show

Checks

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

Commit Message

Jens Freimann Feb. 1, 2019, 10:03 a.m.
Port configuration fails because offload flags don't match the expected
value when max-pkt-len is set to a value that should enable receive port
offloading but doesn't.

There are two cases to consider:

1. VIRTIO_NET_F_MTU is set. Then we need to check if the requested
   max-pkt-len fits into the MTU plus header. If yes we set the
   offload flag.
2. VIRTIO_NET_F_MTU is not set. We can set the offload flag.

Fixes: a4996bd89c42 ("ethdev: new Rx/Tx offloads API")
Cc: stable@dpdk.org

Signed-off-by: Jens Freimann <jfreimann@redhat.com>
---
v3->v4:
 * make use of hw-max_mtu which is calculated during device init. This
   way we don't have to re-read the mtu from config space again. This
   is safe because hw->max_mtu is not changed after init.

v2->v3:
 * remove unnecessary brackets (Maxime)
 * fix commit message (David)

v1->v2:
 * include virtnet hdr, ethernet header, vlan tag when comparing against
   max-rx-pkt-len (Maxime)

 drivers/net/virtio/virtio_ethdev.c | 10 ++++++++++
 1 file changed, 10 insertions(+)

Comments

Ilya Maximets Feb. 1, 2019, 2:27 p.m. | #1
On 01.02.2019 13:03, Jens Freimann wrote:
> Port configuration fails because offload flags don't match the expected
> value when max-pkt-len is set to a value that should enable receive port
> offloading but doesn't.
> 
> There are two cases to consider:
> 
> 1. VIRTIO_NET_F_MTU is set. Then we need to check if the requested
>    max-pkt-len fits into the MTU plus header. If yes we set the
>    offload flag.
> 2. VIRTIO_NET_F_MTU is not set. We can set the offload flag.
> 
> Fixes: a4996bd89c42 ("ethdev: new Rx/Tx offloads API")
> Cc: stable@dpdk.org
> 
> Signed-off-by: Jens Freimann <jfreimann@redhat.com>
> ---
> v3->v4:
>  * make use of hw-max_mtu which is calculated during device init. This
>    way we don't have to re-read the mtu from config space again. This
>    is safe because hw->max_mtu is not changed after init.
> 
> v2->v3:
>  * remove unnecessary brackets (Maxime)
>  * fix commit message (David)
> 
> v1->v2:
>  * include virtnet hdr, ethernet header, vlan tag when comparing against
>    max-rx-pkt-len (Maxime)
> 
>  drivers/net/virtio/virtio_ethdev.c | 10 ++++++++++
>  1 file changed, 10 insertions(+)
> 
> diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
> index 7c4c1df00..4cf91909d 100644
> --- a/drivers/net/virtio/virtio_ethdev.c
> +++ b/drivers/net/virtio/virtio_ethdev.c
> @@ -2351,6 +2351,16 @@ virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
>  	if ((host_features & tso_mask) == tso_mask)
>  		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
>  
> +	if (host_features & (1ULL << VIRTIO_NET_F_MTU)) {
> +		uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
> +			hw->vtnet_hdr_size;
> +		if (dev->data->dev_conf.rxmode.max_rx_pkt_len <=
> +				hw->max_mtu + ether_hdr_len)
> +			dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
> +	} else {
> +		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
> +	}
> +

As I wrote for v3, hw->max_mtu already calculated taking VIRTIO_NET_F_MTU
into account. If VIRTIO_NET_F_MTU is not set, hw->max_mtu is equal to
"VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN - VLAN_TAG_LEN - hw->vtnet_hdr_size".
i.e. "hw->max_mtu + ether_hdr_len" equal to VIRTIO_MAX_RX_PKTLEN which is
larger or equal to rxmode.max_rx_pkt_len.
So, there is no need to check for VIRTIO_NET_F_MTU here. You may just perform
same check for both cases.

This doesn't give any performance or so, but will simplify the code.

>  	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
>  				    DEV_TX_OFFLOAD_VLAN_INSERT;
>  	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {
>
Jens Freimann Feb. 1, 2019, 3:13 p.m. | #2
On Fri, Feb 01, 2019 at 05:27:37PM +0300, Ilya Maximets wrote:
>On 01.02.2019 13:03, Jens Freimann wrote:
>> +	if (host_features & (1ULL << VIRTIO_NET_F_MTU)) {
>> +		uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
>> +			hw->vtnet_hdr_size;
>> +		if (dev->data->dev_conf.rxmode.max_rx_pkt_len <=
>> +				hw->max_mtu + ether_hdr_len)
>> +			dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
>> +	} else {
>> +		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
>> +	}
>> +
>
>As I wrote for v3, hw->max_mtu already calculated taking VIRTIO_NET_F_MTU
>into account. If VIRTIO_NET_F_MTU is not set, hw->max_mtu is equal to
>"VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN - VLAN_TAG_LEN - hw->vtnet_hdr_size".
>i.e. "hw->max_mtu + ether_hdr_len" equal to VIRTIO_MAX_RX_PKTLEN which is
>larger or equal to rxmode.max_rx_pkt_len.
>So, there is no need to check for VIRTIO_NET_F_MTU here. You may just perform
>same check for both cases.

I should read more carefully :). I think I get what you mean now.
hw->max_mtu already includes ether_hdr_len, so it doesn't need
re-calculation here. And in case of VIRTIO_NET_F_MTU the mtu is
checked during feature negotiation.  

So it basically boils down to 
        if (dev->data->dev_conf.rxmode.max_rx_pkt_len <= w->max_mtu)
                dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;

>This doesn't give any performance or so, but will simplify the code.

Thanks for the review!

regards,
Jens
Ilya Maximets Feb. 1, 2019, 3:21 p.m. | #3
On 01.02.2019 18:13, Jens Freimann wrote:
> On Fri, Feb 01, 2019 at 05:27:37PM +0300, Ilya Maximets wrote:
>> On 01.02.2019 13:03, Jens Freimann wrote:
>>> +    if (host_features & (1ULL << VIRTIO_NET_F_MTU)) {
>>> +        uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
>>> +            hw->vtnet_hdr_size;
>>> +        if (dev->data->dev_conf.rxmode.max_rx_pkt_len <=
>>> +                hw->max_mtu + ether_hdr_len)
>>> +            dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
>>> +    } else {
>>> +        dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
>>> +    }
>>> +
>>
>> As I wrote for v3, hw->max_mtu already calculated taking VIRTIO_NET_F_MTU
>> into account. If VIRTIO_NET_F_MTU is not set, hw->max_mtu is equal to
>> "VIRTIO_MAX_RX_PKTLEN - ETHER_HDR_LEN - VLAN_TAG_LEN - hw->vtnet_hdr_size".
>> i.e. "hw->max_mtu + ether_hdr_len" equal to VIRTIO_MAX_RX_PKTLEN which is
>> larger or equal to rxmode.max_rx_pkt_len.
>> So, there is no need to check for VIRTIO_NET_F_MTU here. You may just perform
>> same check for both cases.
> 
> I should read more carefully :). I think I get what you mean now.
> hw->max_mtu already includes ether_hdr_len, so it doesn't need
> re-calculation here. And in case of VIRTIO_NET_F_MTU the mtu is
> checked during feature negotiation. 
> So it basically boils down to        if (dev->data->dev_conf.rxmode.max_rx_pkt_len <= w->max_mtu)
>                dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
> 

'hw->max_mtu' does not include 'ether_hdr_len'. But yes, VIRTIO_NET_F_MTU
already checked during negotiation. So, it will be:

if (dev->data->dev_conf.rxmode.max_rx_pkt_len <= hw->max_mtu + ether_hdr_len)
    dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;

>> This doesn't give any performance or so, but will simplify the code.
> 
> Thanks for the review!
> 
> regards,
> Jens
>

Patch

diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
index 7c4c1df00..4cf91909d 100644
--- a/drivers/net/virtio/virtio_ethdev.c
+++ b/drivers/net/virtio/virtio_ethdev.c
@@ -2351,6 +2351,16 @@  virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 	if ((host_features & tso_mask) == tso_mask)
 		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
 
+	if (host_features & (1ULL << VIRTIO_NET_F_MTU)) {
+		uint32_t ether_hdr_len = ETHER_HDR_LEN + VLAN_TAG_LEN +
+			hw->vtnet_hdr_size;
+		if (dev->data->dev_conf.rxmode.max_rx_pkt_len <=
+				hw->max_mtu + ether_hdr_len)
+			dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
+	} else {
+		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_JUMBO_FRAME;
+	}
+
 	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
 				    DEV_TX_OFFLOAD_VLAN_INSERT;
 	if (host_features & (1ULL << VIRTIO_NET_F_CSUM)) {