[ovs-dev] [PATCH ovs V1 2/9] dpif-hw-acc: Add tc interface

[Paul Blakey]

On 16/11/2016 20:59, Joe Stringer wrote:
>>>> +
>>>> +
>>>> +bool SKIP_HW = false;
>>> What is the granularity of this configuration option supposed to be?
>>
>> This variable corresponds to flower classifier skip_hw/skip_sw flags, so
>> the granularity would have been true/false. Right now this variable changes
>> according to the bridge name. We thought about moving this to vsctl
>> other_config variable.
>> Its purpose was to compare HW offload  vs handling in tc/software.
> By granularity, I meant how widely is it supposed to apply? It's
> currently set in dpif_hw_acc_open() and port_add() - and only ever set
> true.
>
> Is this parameter intended to be specified in the database by the
> user, or is it more of a debug thing?
Initially it was a debug thing. It defaults to false (the non debug mode 
- HW offload)
and changing it to true was done by creating a bridge named skip_hw 
(which adds a internal port named the same).
We didn't bother having a way to return it to false since its for debug, 
we just restarted the daemon.
Wasn't the most graceful method.

It was found to be very useful, so its going to be specified in the 
database.