Re: [dpdk-dev] [PATCH v3] ethdev: fix DCB config issue on ixgbe

["Lu, Wenzhuo" <wenzhuo.lu@intel.com>]

Hi Thomas,


> -----Original Message-----
> From: Lu, Wenzhuo
> Sent: Thursday, June 30, 2016 4:23 PM
> To: 'Thomas Monjalon'
> Cc: dev@dpdk.org
> Subject: RE: [dpdk-dev] [PATCH v3] ethdev: fix DCB config issue on ixgbe
> 
> Hi Thomas,
> 
> > -----Original Message-----
> > From: Thomas Monjalon [mailto:thomas.monjalon@6wind.com]
> > Sent: Thursday, June 30, 2016 3:42 PM
> > To: Lu, Wenzhuo
> > Cc: dev@dpdk.org
> > Subject: Re: [dpdk-dev] [PATCH v3] ethdev: fix DCB config issue on
> > ixgbe
> >
> > 2016-06-30 01:40, Lu, Wenzhuo:
> > > From: dev [mailto:dev-bounces@dpdk.org] On Behalf Of Lu, Wenzhuo
> > > > From: Thomas Monjalon [mailto:thomas.monjalon@6wind.com]
> > > > > 2016-06-23 01:04, Lu, Wenzhuo:
> > > > > > From: Thomas Monjalon [mailto:thomas.monjalon@6wind.com]
> > > > > > > 2016-05-06 05:33, Wenzhuo Lu:
> > > > > > > > +int
> > > > > > > > +rte_eth_dev_mq_mode_set(uint8_t port_id,
> > > > > > > > +			enum rte_eth_rx_mq_mode rx_mq_mode,
> > > > > > > > +			enum rte_eth_tx_mq_mode tx_mq_mode);
> > > > > > >
> > > > > > > I've really tried to think about it and I think it is more or less a hack.
> > > > > > > First, it is not explained in the doc when we should use
> > > > > > > rte_eth_dev_mq_mode_set() instead of a simple call to
> > > > > rte_eth_dev_configure().
> > > > > > > Second, I don't understand why having a function which
> > > > > > > configures the "multiqueue modes" without configuring
> > > > > > > properly
> > RSS/VMDq/DCB.
> > > > > > > Last, it is said that rte_eth_dev_configure() "must be
> > > > > > > invoked first before any other function in the Ethernet API".
> > > After checking the code, Honestly I'm confused. I don't find this description.
> >
> > It's in the description of rte_eth_dev_configure():
> > 	http://dpdk.org/browse/dpdk/tree/lib/librte_ether/rte_ethdev.h#n1904
> The bad news is this rule is not followed. I think the reason is something has to
> be done before configure.
> 
> >
> > > And on the contrary, I find rte_eth_dev_info_get is always called
> > > before rte_eth_dev_configure. I believe it's the problem.
> > > As rte_eth_dev_configure is not run, rte_eth_dev_info_get cannot get
> > > the
> > right info.
> > > That why I have to add a API to set the mq_mode before
> > rte_eth_dev_info_get.
> > > Does that mean all the related examples are wrong? Any opinion? Thanks.
> >
> > My opinion is that this area needs a good cleanup and easy API :)
> OK. My solution is
> 1, Remove the description " rte_eth_dev_configure() must be invoked first
> before any other function in the Ethernet API ". I think it's reasonable to execute
> the rte_eth_dev_info_get before rte_eth_dev_configure, because we need to
> get some NIC specific limitation to check if the configuration is right.
> 2, Add one more argument, " const struct rte_eth_conf *eth_conf ", for
> rte_eth_dev_info_get, like this, void rte_eth_dev_info_get(uint8_t port_id,
> struct rte_eth_dev_info *dev_info, const struct rte_eth_conf *eth_conf);
> because getting the right info depends on the configuration.
And I'm a little lost, my patch has nothing to do with rte_eth_dev_configure. I just try to fix the issue that rte_eth_dev_info_get has dependency with configuration. If the configure is not right, we cannot get the right info.