Re: [dpdk-dev] [PATCH v1 06/11] ixgbe: fix rx intr compatible issue with PF mbox

["Ananyev, Konstantin" <konstantin.ananyev@intel.com>]

 
> 
> From: David Marchand [mailto:david.marchand@6wind.com]
> Sent: Monday, November 02, 2015 4:22 PM
> To: Ananyev, Konstantin
> Cc: Liang, Cunming; dev@dpdk.org
> Subject: Re: [dpdk-dev] [PATCH v1 06/11] ixgbe: fix rx intr compatible issue with PF mbox
> 
> On Mon, Nov 2, 2015 at 5:09 PM, Ananyev, Konstantin <konstantin.ananyev@intel.com> wrote:
> 
> > -----Original Message-----
> > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of David Marchand
> > Sent: Monday, November 02, 2015 4:03 PM
> > To: Liang, Cunming
> > Cc: dev@dpdk.org
> > Subject: Re: [dpdk-dev] [PATCH v1 06/11] ixgbe: fix rx intr compatible issue with PF mbox
> >
> > On Thu, Sep 24, 2015 at 7:33 AM, Cunming Liang <cunming.liang@intel.com>
> > wrote:
> >
> > > When ixgbe runs as a PF, mbox interrupt is prerequisite to make VF start
> > > normally.
> > > And PF sometimes won't 'dev_start', so the mbox interrupt register during
> > > 'dev_init' is required.
> > >
> >
> > Can you describe the cases/situations where you would want a device to
> > handle interrupts while not started ?
> 
> When PF and VF are both controlled by DPDK process(es).
> And user doesn't really want to do any RX/TX through PF - uses PF just to control/configure VF(s).
> 
> Ok, but the user still needs to whitelist the PF (or ensure the PF is not blacklisted) in one of these processes.

Yes, at least dev_init() need to be called for that device. 

> Then, the application would do a "partial" initialisation ?

Yep, sort of.

> If you don't want rx/tx, don't poll the port.

Well, the question is why to add an extra restriction here?
Probably user deliberately doesn't want to call dev_start() for PF device -
as he doesn't plan to use it for RX/TX.
Or might be dev_stop() was called just to do some re-configuration 
(allow to TX scattered packets on the PF queues or so).
Or dev_start() for PF has not yet been called.
Why VF should stop working properly because of that? 
Konstantin

> Anyway, this is your code :-)
> 
> 
> --
> David Marchand