broadcast

[smolkaj]

The architecture v1model.p4 does not currently seem to provide a primitive to broadcast packets.

[mihaibudiu]

As far as I know there is no such primitive supported in either P4-14 or the behavioral model at the level of P4.
How multicast is done is a property of the architecture. If I recall correctly, in the behavioral model this is done by setting some special metadata fields and populating a hidden magic table from the control plane.

[smolkaj]

Would it be better to model this as an extern in P4-16? Say

extern broadcast(standard_metadata_t meta);

? There is already a similar extern

extern void mark_to_drop();

in v1model.p4 (which, on a side node, might be better modeled as

extern void mark_to_drop(standard_metadata_t meta)

).

[mihaibudiu]

We don't have much freedom in writing the v1model.p4, it has to match closely the way P4-14 works, including its switch model, and it also has to be quite close to the behavioral model. The only reason for the existence of v1model is to preserve backward compatibility with P4-14. The drop action from P4-14 is translated to mark_to_drop() - a more suitable name. This is probably not the same as setting the drop bit in the standard_metadata (you have to read the P4-14 spec to know what that bit does). There is no broadcast in P4-14.

We should not be writing P4-16 code for v1model.p4 at all, but since we don't have a standard model and a supporting behavioral model implementation, this is currently the only way to test the compiler end-to-end.

So perhaps you are suggesting to add "broadcast" to the standard model.

[smolkaj]

@mbudiu-vmw: I see, that makes sense.