[dpdk-dev] [PATCH v9 3/4] bus/vdev: move to vdev bus to drivers/bus

[Gaëtan Rivet]

On Fri, Oct 27, 2017 at 10:53:11AM +0200, Thomas Monjalon wrote:
> 27/10/2017 10:19, Tan, Jianfeng:
> > From: Thomas Monjalon [mailto:thomas at monjalon.net]
> > > 27/10/2017 03:06, Jianfeng Tan:
> > > > --- /dev/null
> > > > +++ b/drivers/bus/vdev/rte_bus_vdev_version.map
> > > > @@ -0,0 +1,8 @@
> > > > +DPDK_17.11 {
> > > > +       global:
> > > > +
> > > > +       rte_vdev_init;
> > > > +       rte_vdev_register;
> > > > +       rte_vdev_uninit;
> > > > +       rte_vdev_unregister;
> > > > +};
> > > 
> > > It should not be needed to export the driver ops.
> > 
> > rte_vdev_register/unregister are needed by vdev PMDs, which are needed to export, isn't it?
> 
> Yes you're right, I overlooked it.
> 
> > And for rte_vdev_init()/rte_vdev_uninit(), do you mean we all change to use rte_eal_dev_attach()/detach()?
> 
> rte_vdev_init() is mostly used in tests.
> I changed my mind, you can keep it.
> Maybe it would be clearer to rename rte_vdev.h to rte_bus_vdev.h
> to mimic what was done for PCI. So rte_bus_*.h are mainly for PMDs
> and not to be used by common applications.
> 
> > > Please can you try to make it private?
> > 
> > Even we want to make it private, I think it needs deprecation notice, which we can further improved for the next release?
> 
> Yes we can revise usage of rte_vdev_init(), rte_eal_dev_attach() and
> rte_eal_hotplug_add() in the next release.
> 
> Gaetan, do you have any comment?

Well, as stated before, I agree that deprecating attach / detach seems
good to me. The hotplug API supersedes it and the only difference from a
user PoV is the artificial restriction to vdev and PCI buses.

-- 
Gaëtan Rivet
6WIND