Re: [PATCH v7 01/23] net: Introduce direct data placement tcp offload

[Jakub Kicinski <kuba@kernel.org>]

On Wed, 26 Oct 2022 15:01:42 +0000 Shai Malin wrote:
> > > @@ -14,7 +14,7 @@ typedef u64 netdev_features_t;
> > >  enum {
> > >       NETIF_F_SG_BIT,                 /* Scatter/gather IO. */
> > >       NETIF_F_IP_CSUM_BIT,            /* Can checksum TCP/UDP over IPv4. */
> > > -     __UNUSED_NETIF_F_1,
> > > +     NETIF_F_HW_ULP_DDP_BIT,         /* ULP direct data placement offload */  
> > 
> > Why do you need a feature bit if there is a whole caps / limit querying
> > mechanism?  
> 
> The caps are used for the HW device to publish the supported 
> capabilities/limitation, while the feature bit is used for the DDP 
> enablement "per net-device".
> 
> Disabling will be required in case that another feature which is 
> mutually exclusive to the DDP is needed (as an example in the mlx case, 
> CQE compress which is controlled from ethtool).

It's a big enough feature to add a genetlink or at least a ethtool
command to control. If you add more L5 protos presumably you'll want
to control disable / enable separately for them. Also it'd be cleaner
to expose the full capabilities and report stats via a dedicated API.
Feature bits are not a good fix for complex control-pathy features.

> > It's somewhat unclear to me why we add ops to struct net_device,
> > rather than to ops.. can you explain?  
> 
> We were trying to follow the TLS design which is similar.

Ack, TLS should really move as well, IMHO, but that's a separate convo.

> Can you please clarify what do you mean by "rather than to ops.."?

Add the ulp_dpp_ops pointer to struct net_device_ops rather than struct
net_device.