From: Tom Herbert <t...@herbertland.com> Date: Fri, 3 Feb 2017 14:25:00 -0800
> On Fri, Feb 3, 2017 at 2:02 PM, David Miller <da...@davemloft.net> wrote: >> From: Tom Herbert <t...@herbertland.com> >> Date: Fri, 3 Feb 2017 13:58:56 -0800 >> >>> On Fri, Feb 3, 2017 at 1:50 PM, David Miller <da...@davemloft.net> wrote: >>>> From: Michael Chan <michael.c...@broadcom.com> >>>> Date: Fri, 3 Feb 2017 13:13:47 -0800 >>>> >>>>> On Fri, Feb 3, 2017 at 12:49 PM, David Miller <da...@davemloft.net> wrote: >>>>>> >>>>>> Please _DO NOT_ guard XDP support with an ifdef the user >>>>>> can modify. >>>>>> >>>>>> Treat it like any other common netdev feature a driver might >>>>>> support such as checksum offloading or GRO. >>>>>> >>>>> >>>>> David, I want to make sure I understand completely. Are you saying >>>>> don't use Kconfig option for XDP? Have it always available? >>>> >>>> Yes. >>>> >>>> I don't see a similar config option used in any other driver. >>>> >>>> What's really driving me completely mad about driver XDP adoption >>>> is that there is so much inconsistency. >>>> >>>> If you do not see another XDP supporting driver do something, don't be >>>> tempted to blaze your own trail and handle something in a unique way. >>>> >>>> We don't set precedence by one driver saying "hey it's better to do >>>> things this way, forget what all the other drivers are doing." Rather >>>> we have a "discussion" about what the appropriate thing is to do and >>>> convert all the drivers only after a decision has been made. >>>> >>>> Meanwhile we keep the status quo. >>> >>> I am working on some API changes that will hopefully get a little >>> consistency across these drivers (this includes feature flag >>> NETIF_F_XDP). This will reduce code some and should be good cleanup, >>> but XDP is currently very intertwined with the critical data path so >>> we might need to be looking at this for a while. There's now more >>> drivers with XDP support than when I started this work, so I don't >>> think bnxt should wait for this cleanup-- it's just one more driver >>> we'll have to retrofit. >> >> Of course. >> >> Michael just respin with the Kconfig change and I'll apply your >> series. In fact I was about to until I noticed the XDP Kconfig knob >> :) > > Meaning no Kconfig and no features flag I assume... For now, yes.