[dpdk-dev] [RFC PATCH] ethdev: remove old flow director API

[Wu, Jingjing]

> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> Sent: Monday, April 20, 2015 11:00 PM
> To: Wu, Jingjing
> Cc: dev at dpdk.org
> Subject: Re: [RFC PATCH] ethdev: remove old flow director API
> 
> 2015-04-20 14:32, Wu, Jingjing:
> > From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> > > It's time to remove this old API.
> > > It seems some work is still needed to rely only on eth_ctrl API.
> > > At least ixgbe, i40e and testpmd must be fixed.
> > > Jingjing, do you think it's possible to remove all these structures
> > > from rte_ethdev.h?
> > >
> > [Wu, Jingjing] Yes, I agree.
> > But few comments list below.
> > Beside the following change, some commands also need to be removed in testpmd. For the
> ixgbe, code to the old APIs are already fixed.
> 
> > > --- a/lib/librte_ether/rte_ethdev.h
> > > +++ b/lib/librte_ether/rte_ethdev.h
> > > @@ -672,42 +672,6 @@ struct rte_eth_pfc_conf {
> > >  };
> > >
> > >  /**
> > > - *  Memory space that can be configured to store Flow Director filters
> > > - *  in the board memory.
> > > - */
> > > -enum rte_fdir_pballoc_type {
> > > -	RTE_FDIR_PBALLOC_64K = 0,  /**< 64k. */
> > > -	RTE_FDIR_PBALLOC_128K,     /**< 128k. */
> > > -	RTE_FDIR_PBALLOC_256K,     /**< 256k. */
> > > -};
> 
> This is hardware-related. Why should it be part of the API?
> 
> > > -
> > > -/**
> > > - *  Select report mode of FDIR hash information in RX descriptors.
> > > - */
> > > -enum rte_fdir_status_mode {
> > > -	RTE_FDIR_NO_REPORT_STATUS = 0, /**< Never report FDIR hash. */
> > > -	RTE_FDIR_REPORT_STATUS, /**< Only report FDIR hash for matching pkts. */
> > > -	RTE_FDIR_REPORT_STATUS_ALWAYS, /**< Always report FDIR hash. */
> > > -};
> > > -
> > > -/**
> > > - * A structure used to configure the Flow Director (FDIR) feature
> > > - * of an Ethernet port.
> > > - *
> > > - * If mode is RTE_FDIR_DISABLE, the pballoc value is ignored.
> > > - */
> > > -struct rte_fdir_conf {
> > > -	enum rte_fdir_mode mode; /**< Flow Director mode. */
> > > -	enum rte_fdir_pballoc_type pballoc; /**< Space for FDIR filters. */
> > > -	enum rte_fdir_status_mode status;  /**< How to report FDIR hash. */
> > > -	/** RX queue of packets matching a "drop" filter in perfect mode. */
> > > -	uint8_t drop_queue;
> > > -	struct rte_eth_fdir_masks mask;
> > > -	struct rte_eth_fdir_flex_conf flex_conf;
> > > -	/**< Flex payload configuration. */
> > > -};
> > > -
> > [Wu, Jingjing] This structures and above types are useful about global configuration. They
> can't be removed.
> 
> Why? I thought eth_ctrl API would be enough.
> It seems strange to have a part of the flow director API in rte_eth_ctrl.h and
> another part in rte_ethdev.h. Other filters have no impact on rte_ethdev.h.

rte_fdir_conf in rte_eth_conf is used when rte_eth_dev_configure to set initialization parameters which cann't be changed at running time.
For example, if we disable the flow director, we can set the mode in it to RTR_FDIR_DISABLE, then the resource will not be allocated for flow director.
The way is the same as rss_conf in rte_eth_conf. While the filter_ctrl can provide interface to configure parameters or add/delete rules at device running time.