[dpdk-dev] [PATCH v2] net/vhost: convert to new Rx/Tx offload API

[Maxime Coquelin]

On 05/22/2018 02:38 PM, Ferruh Yigit wrote:
> On 5/22/2018 1:34 PM, Tiwei Bie wrote:
>> On Tue, May 22, 2018 at 02:22:11PM +0200, Maxime Coquelin wrote:
>>> Signed-off-by: Maxime Coquelin <maxime.coquelin at redhat.com>
>>> ---
>>> Changes since v2:
>>> - Announce supported Rx & Tx offload features (Tiwei)
>>>
>>>   drivers/net/vhost/rte_eth_vhost.c | 8 ++++++--
>>>   1 file changed, 6 insertions(+), 2 deletions(-)
>>>
>>> diff --git a/drivers/net/vhost/rte_eth_vhost.c b/drivers/net/vhost/rte_eth_vhost.c
>>> index f473bbbb3..229a5ba8d 100644
>>> --- a/drivers/net/vhost/rte_eth_vhost.c
>>> +++ b/drivers/net/vhost/rte_eth_vhost.c
>>> @@ -491,9 +491,9 @@ eth_dev_configure(struct rte_eth_dev *dev __rte_unused)
>>>   	struct pmd_internal *internal = dev->data->dev_private;
>>>   	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
>>>   
>>> -	internal->vlan_strip = rxmode->hw_vlan_strip;
>>> +	internal->vlan_strip = !!(rxmode->offloads & DEV_RX_OFFLOAD_VLAN_STRIP);
>>>   
>>> -	if (rxmode->hw_vlan_filter)
>>> +	if (rxmode->offloads & DEV_RX_OFFLOAD_VLAN_FILTER)
>>>   		VHOST_LOG(WARNING,
>>>   			"vhost(%s): vlan filtering not available\n",
>>>   			internal->dev_name);
>>
>> I'm not quite sure whether we still need to log this warning.
> 
> Similar warning exists in ethdev layer, when an offload requested that is not
> reported in capability. So I agree to remove this.
> 
> And in next release that warning will be converted into error, so that PMD will
> not get those values at all.

Ok, I'll post a v3 removing the check.

Thanks,
Maxime

>>
>>> @@ -1072,6 +1072,10 @@ eth_dev_info(struct rte_eth_dev *dev,
>>>   	dev_info->max_rx_queues = internal->max_queues;
>>>   	dev_info->max_tx_queues = internal->max_queues;
>>>   	dev_info->min_rx_bufsize = 0;
>>> +
>>> +	dev_info->tx_offload_capa = DEV_TX_OFFLOAD_MULTI_SEGS |
>>> +				DEV_TX_OFFLOAD_VLAN_INSERT;
>>> +	dev_info->rx_offload_capa = DEV_RX_OFFLOAD_VLAN_STRIP;
>>>   }
>>
>> Other than that, it looks good to me. Thanks!
>>
>> Best regards,
>> Tiwei Bie
>>
>