Re: [PATCH net-next] sfc: support offloading TC VLAN push/pop actions to the MAE

[Martin Habets <habetsm.xilinx@gmail.com>]

On Tue, Feb 21, 2023 at 08:32:13PM +0000, Edward Cree wrote:
> On 19/02/2023 09:21, Leon Romanovsky wrote:
> > On Thu, Feb 16, 2023 at 04:04:42PM +0000, edward.cree@amd.com wrote:
> >> From: Edward Cree <ecree.xilinx@gmail.com>
> >>
> >> EF100 can pop and/or push up to two VLAN tags.
> >>
> >> Signed-off-by: Edward Cree <ecree.xilinx@gmail.com>
> ...
> >> +	/* Translate vlan actions from bitmask to count */
> >> +	switch (act->vlan_push) {
> >> +	case 0:
> >> +	case 1:
> >> +		vlan_push = act->vlan_push;
> >> +		break;
> >> +	case 2: /* can't happen */
> > 
> > There is no need in case here as "default" will catch.
> > 
> >> +	default:
> >> +		return -EINVAL;
> >> +	case 3:
> >> +		vlan_push = 2;
> >> +		break;
> >> +	}
> >> +	switch (act->vlan_pop) {
> >> +	case 0:
> >> +	case 1:
> >> +		vlan_pop = act->vlan_pop;
> >> +		break;
> >> +	case 2: /* can't happen */
> >> +	default:
> >> +		return -EINVAL;
> > 
> > Please rely switch-case semantics and don't put default in the middle.
> 
> It's legal C and as far as I can tell there's nothing in coding-style.rst
>  about it; I did it this way so as to put the cases in the logical(?)
>  ascending order and try to make the code self-document the possible
>  values of the act-> fields.
> Arguably it's the 'default:' rather than the 'case 2:' that's unnecessary
>  as the switch argument is an unsigned:2 bitfield, so it can only take on
>  these four values.

Can you replace the switch statement with
 vlan_push = act->vlan_push & 1 + act->vlan_push & 2;
Even then it would seem prudent to guard against  act->vlan_push == 2.

Martin

> Although on revisiting this code I wonder if it makes more sense just to
>  use the 'count' (rather than 'bitmask') form throughout, including in
>  act->vlan_push/pop; it makes the tc.c side of the code slightly more
>  involved, but gets rid of this translation entirely.  WDYT?
> 
> -ed