Darren Reed wrote:
> Folks,
>
> In various places, we have a call, ire_to_ill() to map an ire to
> the appropriate ill for outgoing packets. For any given ire, the
> choice of the ill should be somewhat static and be associated
> with the ire_stq. Given that we already have ire_in_ill, is there
> any reason not to also have ire_out_ill? Or is this optimisation
> not worth the trouble? Furthermore, if we maintain both an
> ire_out_ill and ire_in_ill field in the ire_t, is there any need for
> there to be a function ire_to_ill()?
With the IP datapath refactoring the plan is to replace ire_stq/ire_rfq
with a single ire_ill.
Currently ire_to_ill() needs special checks to handle lo0 (which is an
ill, but doesn't have any corresponding queue_t).
I don't view this as an optimization but as a complexity reduction.
Sebastian wrote:
> The logic currently in ire_to_ill() needs to live somewhere, if only to
> initially set ire_ill. It might as well stay contained in one function.
That isn't true for the lo0 case; we already know the ill_t when the lo0
IRE_INTERFACE is created, but we don't have a queue_t to pin that on.
Erik
_______________________________________________
networking-discuss mailing list
[email protected]
