Re: [dpdk-dev] [PATCH] app/testpmd: change port detach interface

[Thomas Monjalon <thomas@monjalon.net>]

15/05/2019 08:52, Nithin Dabilpuram:
> Hi Thomas,
> On Tue, May 14, 2019 at 05:39:30PM +0200, Thomas Monjalon wrote:
> > Hi,
> > 
> > 13/05/2019 13:21, Nithin Dabilpuram:
> > > With the latest published interface of
> > > rte_eal_hotplug_[add,remove](), and rte_eth_dev_close(),
> > > rte_eth_dev_close() would cleanup all the data structures of
> > > port's eth dev leaving the device common resource intact
> > > if RTE_ETH_DEV_CLOSE_REMOVE is set in dev flags.
> > > So "port detach" (~hotplug remove) should be able to work,
> > > with device identifier like "port attach" as eth_dev could have
> > > been closed already and rte_eth_devices[port_id] reused.
> > 
> > "port attach" uses devargs as identifier because there
> > is no port id before creating it. But "detach port" uses
> > logically the port id to close.
> 
> But if "port close" was already called on that port,
> eth_dev->state would be set as RTE_ETH_DEV_UNUSED and
> that port id could be reused.
> So after "port close" if we call "port detach", isn't it
> incorrect to use the same port id ?

Yes it is incorrect to close a port which is already closed :)

> > > This change alters "port detach" cmdline interface to
> > > work with device identifier like "port attach".
> > 
> > The word "port" means an ethdev port, so it should be
> > referenced with a port id.
> > If you want to close an EAL rte_device, then you should
> > rename the command.
> > But testpmd purpose should be to work with ethdev ports only.
> 
> Renaming the command to "detach <identifier>" ?

Yes something like that.
But why do you want to manage rte_device in testpmd?
Being able to close ports in not enough?
Please describe a scenario.