Re: [PATCH net-next 1/2 RESEND v3] net: use NETIF_F_ALL_TSO for vlan features

Re: [PATCH net-next 1/2 RESEND v3] net: use NETIF_F_ALL_TSO for vlan features

[Dimitris Michailidis]

On 05/10/2011 11:24 PM, Shan Wei wrote:
> As Dimitris Michailidis suggested, use NETIF_F_ALL_TSO for vlan_features,
> which is a mask, but not hw_features.
> 
> Signed-off-by: Shan Wei <shanwei@cn.fujitsu.com>

While these changes aren't wrong I don't see a good reason to make them.  I am 
also curious why you're changing only these three drivers.

> ---
>  drivers/net/bnx2x/bnx2x_main.c     |    2 +-
>  drivers/net/cxgb4/cxgb4_main.c     |    2 +-
>  drivers/net/cxgb4vf/cxgb4vf_main.c |    2 +-
>  3 files changed, 3 insertions(+), 3 deletions(-)
> 
> diff --git a/drivers/net/bnx2x/bnx2x_main.c b/drivers/net/bnx2x/bnx2x_main.c
> index 2762edf..172684b 100644
> --- a/drivers/net/bnx2x/bnx2x_main.c
> +++ b/drivers/net/bnx2x/bnx2x_main.c
> @@ -9267,7 +9267,7 @@ static int __devinit bnx2x_init_dev(struct pci_dev *pdev,
>  		NETIF_F_RXCSUM | NETIF_F_LRO | NETIF_F_HW_VLAN_TX;
>  
>  	dev->vlan_features = NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
> -		NETIF_F_TSO | NETIF_F_TSO_ECN | NETIF_F_TSO6 | NETIF_F_HIGHDMA;
> +		NETIF_F_ALL_TSO | NETIF_F_HIGHDMA;
>  
>  	dev->features |= dev->hw_features | NETIF_F_HW_VLAN_RX;
>  	if (bp->flags & USING_DAC_FLAG)
> diff --git a/drivers/net/cxgb4/cxgb4_main.c b/drivers/net/cxgb4/cxgb4_main.c
> index 7e3cfbe..21a163c 100644
> --- a/drivers/net/cxgb4/cxgb4_main.c
> +++ b/drivers/net/cxgb4/cxgb4_main.c
> @@ -3528,7 +3528,7 @@ static void free_some_resources(struct adapter *adapter)
>  }
>  
>  #define TSO_FLAGS (NETIF_F_TSO | NETIF_F_TSO6 | NETIF_F_TSO_ECN)
> -#define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | TSO_FLAGS | \
> +#define VLAN_FEAT (NETIF_F_SG | NETIF_F_IP_CSUM | NETIF_F_ALL_TSO | \
>  		   NETIF_F_IPV6_CSUM | NETIF_F_HIGHDMA)
>  
>  static int __devinit init_one(struct pci_dev *pdev,
> diff --git a/drivers/net/cxgb4vf/cxgb4vf_main.c b/drivers/net/cxgb4vf/cxgb4vf_main.c
> index e71c08e..b55c521 100644
> --- a/drivers/net/cxgb4vf/cxgb4vf_main.c
> +++ b/drivers/net/cxgb4vf/cxgb4vf_main.c
> @@ -2604,7 +2604,7 @@ static int __devinit cxgb4vf_pci_probe(struct pci_dev *pdev,
>  		netdev->hw_features = NETIF_F_SG | TSO_FLAGS |
>  			NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
>  			NETIF_F_HW_VLAN_TX | NETIF_F_RXCSUM;
> -		netdev->vlan_features = NETIF_F_SG | TSO_FLAGS |
> +		netdev->vlan_features = NETIF_F_SG | NETIF_F_ALL_TSO |
>  			NETIF_F_IP_CSUM | NETIF_F_IPV6_CSUM |
>  			NETIF_F_HIGHDMA;
>  		netdev->features = netdev->hw_features |

Re: [PATCH net-next 1/2 RESEND v3] net: use NETIF_F_ALL_TSO for vlan features

[Shan Wei]

Dimitris Michailidis wrote, at 2011年05月12日 00:35:
> On 05/10/2011 11:24 PM, Shan Wei wrote:
>> As Dimitris Michailidis suggested, use NETIF_F_ALL_TSO for vlan_features,
>> which is a mask, but not hw_features.
>>
>> Signed-off-by: Shan Wei <shanwei@cn.fujitsu.com>
> 
> While these changes aren't wrong I don't see a good reason to make them.  I am also curious why you're changing only these three drivers.

A personal reason to quickly know what device supports ALL TSO,
not to search one by one. :;)

TSO offloads are assigned directly in these three drivers.
Other drivers set vlan_features as hw_features, or set TSO offloads after doing some check.

-- 
Best Regards
-----
Shan Wei

Re: [PATCH net-next 1/2 RESEND v3] net: use NETIF_F_ALL_TSO for vlan features

[Dimitris Michailidis]

On 05/12/2011 02:54 AM, Shan Wei wrote:
> Dimitris Michailidis wrote, at 2011年05月12日 00:35:
>> On 05/10/2011 11:24 PM, Shan Wei wrote:
>>> As Dimitris Michailidis suggested, use NETIF_F_ALL_TSO for vlan_features,
>>> which is a mask, but not hw_features.
>>>
>>> Signed-off-by: Shan Wei <shanwei@cn.fujitsu.com>
>> While these changes aren't wrong I don't see a good reason to make them.  I am also curious why you're changing only these three drivers.
> 
> A personal reason to quickly know what device supports ALL TSO,
> not to search one by one. :;)

But vlan_features doesn't indicate by itself what the device supports, and it 
is because of this that you can use NETIF_F_ALL_TSO in vlan_features without 
much future breakage risk.  There's still the potential that in the future some 
additional TSO variant will be added that some of the drivers you're changing 
support, but not over VLANs, and then they'll need to undo you change.

If you want to know which drivers support all TSOs you really need to look at 
their features and hw_features.  One could set NETIF_F_ALL_TSO in vlan_features 
of a driver that doesn't support any form of TSO.  The way you're checking 
you'd be misled into thinking that driver supports all TSO.

Re: [PATCH net-next 1/2 RESEND v3] net: use NETIF_F_ALL_TSO for vlan features

[Shan Wei]

Dimitris Michailidis wrote, at 2011年05月13日 01:05:
> But vlan_features doesn't indicate by itself what the device supports, and it is because of this that you can use NETIF_F_ALL_TSO in vlan_features without much future breakage risk.  There's still the potential that in the future some additional TSO variant will be added that some of the drivers you're changing support, but not over VLANs, and then they'll need to undo you change.
 
For this part.

David, please discard this patch.
 
-- 
Best Regards
-----
Shan Wei

Re: [PATCH net-next 1/2 RESEND v3] net: use NETIF_F_ALL_TSO for vlan features

[David Miller]

From: Shan Wei <shanwei@cn.fujitsu.com>
Date: Fri, 13 May 2011 10:10:21 +0800

> Dimitris Michailidis wrote, at 2011年05月13日 01:05:
>> But vlan_features doesn't indicate by itself what the device supports, and it is because of this that you can use NETIF_F_ALL_TSO in vlan_features without much future breakage risk.  There's still the potential that in the future some additional TSO variant will be added that some of the drivers you're changing support, but not over VLANs, and then they'll need to undo you change.
>  
> For this part.
> 
> David, please discard this patch.

Shan I've discarded all of your patches in this series.