[dpdk-dev] [PATCH 2/2] ethtool: add new library to provide ethtool-alike APIs

[Wang, Liang-min]

> -----Original Message-----
> From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> Sent: Friday, June 05, 2015 9:41 AM
> To: Wang, Liang-min
> Cc: Andrew Harvey (agh); Stephen Hemminger; dev at dpdk.org
> Subject: Re: [dpdk-dev] [PATCH 2/2] ethtool: add new library to provide
> ethtool-alike APIs
> 
> 2015-06-05 11:25, Wang, Liang-min:
> > From: Thomas Monjalon [mailto:thomas.monjalon at 6wind.com]
> > > Stephen and me say the same thing about using the ethdev API.
> > > We don't understand why using a fake ethtool lib would be easier.
> > > Though you are saying it "facilitated [your] adoption to dpdk".
> > > Please could you explain why using an ethtool-like API is easier
> > > than using the existing ethdev API?
> > > In any case, you have to develop a specific backend for DPDK
> > > (rte_ethtool would be also DPDK-specific).
> >
> > As described earlier in this patch comment reply, there are other ethtool
> ops that have been implemented.
> > Those ops includes set/get eeprom, set/get pauseparam, set/get
> ringparam which are not available in the exiting ethdev library.
> 
> 1/ We cannot really consider code which is not public 2/ You may extend
> ethdev if some functions are missing
> 
> > For this release, we focus on releasing some basic functions (btw,
> mac_addr_set is not available but is covered by this patch).
> 
> Yes, you are extending ethdev by adding
> rte_eth_dev_default_mac_addr_set.
> 
> > The key reason that this set of library is not released as part of ethdev is
> the ethtool API dependency on kernel include file.
> 
> It is a good reason to separate the library.
> But it doesn't justify its need.
> 
> > To faithfully carry the ethtool ops and net dev ops API parameters, the
> ethtool APIs are designed to follow the original definition except avoiding
> carry kernel states.
> > With that, to support ethtool APIs faithfully, we need to include
> <linux/ethtool.h>.
> > As suggested by many DPDK veterans including Thomas (indicated over
> your reply), you would prefer these APIs in a separate library.
> 
> I think I'm starting to understand that you really need ethtool conversion
> (implemented in rte_ethtool_get_drvinfo) but not the other functions which
> are simple wrappers. Right?

The rte_ethtool_get_drvinfo and many others ethtool ops have the same conversion requirement.
As for ethtool and net dev ops that don't require conversion. For the sake of clean API interface, they are implemented in the same ethtool library.