[dpdk-dev] [PATCH] event/rx_adapter: fix ignore return of event start

[Rao, Nikhil]

> -----Original Message-----
> From: Varghese, Vipin
> Sent: Friday, February 2, 2018 1:39 PM
> To: Jerin Jacob <jerin.jacob at caviumnetworks.com>; Rao, Nikhil
> <nikhil.rao at intel.com>
> Cc: dev at dpdk.org; Jacob, Jerin <Jerin.JacobKollanukkaran at cavium.com>; Van
> Haaren, Harry <harry.van.haaren at intel.com>; Hemant Agrawal
> <hemant.agrawal at nxp.com>; Jain, Deepak K <deepak.k.jain at intel.com>
> Subject: RE: [PATCH] event/rx_adapter: fix ignore return of event start
> 
> 
> 
> > -----Original Message-----
> > From: Jerin Jacob [mailto:jerin.jacob at caviumnetworks.com]
> > Sent: Wednesday, January 31, 2018 6:54 AM
> > To: Rao, Nikhil <nikhil.rao at intel.com>
> > Cc: Varghese, Vipin <vipin.varghese at intel.com>; dev at dpdk.org; Jacob,
> > Jerin <Jerin.JacobKollanukkaran at cavium.com>; Van Haaren, Harry
> > <harry.van.haaren at intel.com>; Hemant Agrawal
> <hemant.agrawal at nxp.com>;
> > Jain, Deepak K <deepak.k.jain at intel.com>
> > Subject: Re: [PATCH] event/rx_adapter: fix ignore return of event
> > start
> >
> > -----Original Message-----
> > >
> > >
> > > Adding eventdev PMD folks for their suggestions on how to handle the
> > > return
> > value from rte_event_dev_start() below.
> > >
> > > > -----Original Message-----
> > > > From: Varghese, Vipin
> > > > Sent: Wednesday, January 31, 2018 4:26 AM
> > > > To: dev at dpdk.org; Rao, Nikhil <nikhil.rao at intel.com>
> > > > Cc: Jain, Deepak K <deepak.k.jain at intel.com>; Varghese, Vipin
> > > > <vipin.varghese at intel.com>
> > > > Subject: [PATCH] event/rx_adapter: fix ignore return of event
> > > > start
> > > >
> > > > Capture the return value for rte_event_dev_start. Return the
> > > > result back to user.
> > > >
> > > > Coverity issue: 257000
> > > > Fixes: 9c38b704d280 ("eventdev: add eth Rx adapter
> > > > implementation")
> > > > Cc: nikhil.rao at intel.com
> > > >
> > > > Signed-off-by: Vipin Varghese <vipin.varghese at intel.com>
> > > > ---
> > > >  lib/librte_eventdev/rte_event_eth_rx_adapter.c | 4 ++--
> > > >  1 file changed, 2 insertions(+), 2 deletions(-)
> > > >
> > > > diff --git a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> > > > b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> > > > index 90106e6..a818bef 100644
> > > > --- a/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> > > > +++ b/lib/librte_eventdev/rte_event_eth_rx_adapter.c
> > > > @@ -603,7 +603,7 @@ static uint16_t gcd_u16(uint16_t a, uint16_t b)
> > > >  		RTE_EDEV_LOG_ERR("failed to configure event dev %u\n",
> > > >  						dev_id);
> > > >  		if (started)
> > > > -			rte_event_dev_start(dev_id);
> > > > +			ret = rte_event_dev_start(dev_id);
> > >
> > > Currently the a non-zero return value at this point signifies an
> > > error returned
> > from rte_event_dev_configure(),  so I suggest that the return value is
> > typecasted to void.
> >
> > If I understand it correctly, Any one of the failure(configure() or
> > start()) should result in bad state. Right?
> > i.e If some reason PMD is not able to start() even after failure
> > configuration() would result in bad state.
> > If so, one option could be combine the error like ret |= operation or
> > create a new logical error in Rx adapter which denotes this new error.
> >
> 
> So do we agree to ACK these changes to get the code fix to the mainline? 

Sorry, if my original email wasn't clear,  if rte_event_dev_configure() returns an error and rte_eventdev_start() returns success that would be a problem, i.e., the fix is incorrect.

Of the 2 options suggested by Jerin - Since ret is not a bitmask  ret |= wouldn't work, if I understand the option correctly . A new error would work.

How about EIO ? and we also update the documentation to indicate that the event device would be in a stopped state if the return code is EIO.

> rework the logic as required?
> 
> > >
> > > >  		return ret;
> > > >  	}
> > > >
> > > > @@ -617,7 +617,7 @@ static uint16_t gcd_u16(uint16_t a, uint16_t b)
> > > >  	conf->event_port_id = port_id;
> > > >  	conf->max_nb_rx = 128;
> > > >  	if (started)
> > > > -		rte_event_dev_start(dev_id);
> > > > +		ret = rte_event_dev_start(dev_id);
> > > This change looks good to me.
> > >
> > > >  	rx_adapter->default_cb_arg = 1;
> > > >  	return ret;
> > > >  }
> > > > --
> > > > 1.9.1
> > >