net/virtio: set offload flag for jumbo frames
Checks
Commit Message
Port configuration fails when offload flags don't match the expected
value, e.g. when max-pkt-len is set to a value that should enable
receive port offloading but doesn't.
This can be triggered by running testpmd e.g. with --max-pkt-len=2000.
It will fail with "Ethdev port_id=0 requested Rx offloads 0x800 doesn't
match Rx offloads capabilities 0x20d in rte_eth_dev_configure()"
To fix this there are two cases to consider:
1. VIRTIO_NET_F_MTU is negotiated. Then we need to check if the
requested max. packet length fits into the MTU. If yes we set the
offload flag.
2. VIRTIO_NET_F_MTU is not negotiated. We can set the offload flag.
Signed-off-by: Jens Freimann <jfreimann@redhat.com>
---
drivers/net/virtio/virtio_ethdev.c | 11 +++++++++++
1 file changed, 11 insertions(+)
Comments
On 1/30/19 4:26 PM, Jens Freimann wrote:
> Port configuration fails when offload flags don't match the expected
> value, e.g. when max-pkt-len is set to a value that should enable
> receive port offloading but doesn't.
>
> This can be triggered by running testpmd e.g. with --max-pkt-len=2000.
> It will fail with "Ethdev port_id=0 requested Rx offloads 0x800 doesn't
> match Rx offloads capabilities 0x20d in rte_eth_dev_configure()"
>
> To fix this there are two cases to consider:
>
> 1. VIRTIO_NET_F_MTU is negotiated. Then we need to check if the
> requested max. packet length fits into the MTU. If yes we set the
> offload flag.
> 2. VIRTIO_NET_F_MTU is not negotiated. We can set the offload flag.
I think we need to Cc: stable and point to the commit introducing the
regression. I think it is the one introducing the use of
DEV_RX_OFFLOAD_JUMBO_FRAME in Virtio PMD.
>
> Signed-off-by: Jens Freimann <jfreimann@redhat.com>
> ---
> drivers/net/virtio/virtio_ethdev.c | 11 +++++++++++
> 1 file changed, 11 insertions(+)
>
> diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
> index 7c4c1df00..e0d6542d4 100644
> --- a/drivers/net/virtio/virtio_ethdev.c
> +++ b/drivers/net/virtio/virtio_ethdev.c
> @@ -2351,6 +2351,17 @@ 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)) {
> + struct virtio_net_config config;
> + vtpci_read_dev_config(hw,
> + offsetof(struct virtio_net_config, mtu),
> + &config.mtu, sizeof(config.mtu));
> + if (dev->data->dev_conf.rxmode.max_rx_pkt_len <= config.mtu)
That is not exactly right, MTU does not take into account the Ethernet
header, the VLAN tag and the virtio-net header.
See virtio_mtu_set() for example.
> + 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)) {
>
On Wed, Jan 30, 2019 at 04:41:55PM +0100, Maxime Coquelin wrote:
>
>
>On 1/30/19 4:26 PM, Jens Freimann wrote:
>>Port configuration fails when offload flags don't match the expected
>>value, e.g. when max-pkt-len is set to a value that should enable
>>receive port offloading but doesn't.
>>
>>This can be triggered by running testpmd e.g. with --max-pkt-len=2000.
>>It will fail with "Ethdev port_id=0 requested Rx offloads 0x800 doesn't
>>match Rx offloads capabilities 0x20d in rte_eth_dev_configure()"
>>
>>To fix this there are two cases to consider:
>>
>>1. VIRTIO_NET_F_MTU is negotiated. Then we need to check if the
>> requested max. packet length fits into the MTU. If yes we set the
>> offload flag.
>>2. VIRTIO_NET_F_MTU is not negotiated. We can set the offload flag.
>
>I think we need to Cc: stable and point to the commit introducing the
>regression. I think it is the one introducing the use of
>DEV_RX_OFFLOAD_JUMBO_FRAME in Virtio PMD.
ok
>>
>>Signed-off-by: Jens Freimann <jfreimann@redhat.com>
>>---
>> drivers/net/virtio/virtio_ethdev.c | 11 +++++++++++
>> 1 file changed, 11 insertions(+)
>>
>>diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
>>index 7c4c1df00..e0d6542d4 100644
>>--- a/drivers/net/virtio/virtio_ethdev.c
>>+++ b/drivers/net/virtio/virtio_ethdev.c
>>@@ -2351,6 +2351,17 @@ 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)) {
>>+ struct virtio_net_config config;
>>+ vtpci_read_dev_config(hw,
>>+ offsetof(struct virtio_net_config, mtu),
>>+ &config.mtu, sizeof(config.mtu));
>>+ if (dev->data->dev_conf.rxmode.max_rx_pkt_len <= config.mtu)
>
>That is not exactly right, MTU does not take into account the Ethernet
>header, the VLAN tag and the virtio-net header.
>
>See virtio_mtu_set() for example.
I wasn't sure about it, will send a v2. Thanks for the review!
regards,
Jens
@@ -2351,6 +2351,17 @@ 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)) {
+ struct virtio_net_config config;
+ vtpci_read_dev_config(hw,
+ offsetof(struct virtio_net_config, mtu),
+ &config.mtu, sizeof(config.mtu));
+ if (dev->data->dev_conf.rxmode.max_rx_pkt_len <= config.mtu)
+ 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)) {