On Thursday 15 December 2016 07:23 PM, Johannes Berg wrote: > >>> I agree. Dynamic switch part is buggy, we can start with not >>> allowing interfaces resulting in dynamic switch. >> >> Does this mean that when bringing up multiple interfaces, users would >> need to figure out the 'magic' order that works? > > I think we need to talk about hardware capabilities at this point. QCA988X does not have capability to configure vif specific decap mode. Encap mode is configurable per packet for all the ath10k based chips so this part should be fine to support per vif configuration. Newer QCA chips like QCA9984, QCA4019, QCA9888 and QCA99X0 supports decap mode configuration per vif. To reduce the complexity, we can probably make per vif encap/decap configuration mandatory to enable ethernet frame format, not sure how this will work with non-QCA capable hardware. > > I was assuming that it would actually be possible to run two interfaces > with different paths here concurrently - is that not true? If that's > not true, then we absolutely _need_ dynamic switching, I agree with > Felix, but then we have a pretty big complication to figure out. But we > can't let this optimisation affect user experience. Sure. Vasanth
