Re: [dpdk-dev] [PATCH v7 1/2] i40e: RSS/FD granularity configuration

[Thomas Monjalon <thomas.monjalon@6wind.com>]

2015-11-02 07:59, Zhang, Helin:
> From: Thomas Monjalon [mailto:thomas.monjalon@6wind.com]
> > 2015-11-02 02:11, Zhang, Helin:
> > > From: Thomas Monjalon [mailto:thomas.monjalon@6wind.com]
> > > > 2015-10-30 11:11, Helin Zhang:
> > > > > v7 changes:
> > > > > Modified RTE_ETH_INSET_SIZE_MAX from 128 to 1, to avoid ABI changes.
> > > >
> > > > Why it would avoid ABI changes?
> > > > Have you passed the ABI checker?
> > >
> > > Hi Thomas
> > >
> > > The only structure modified is ' struct rte_eth_hash_filter_info ',
> > > where a new element (struct rte_eth_input_set_conf input_set_conf) was
> > > added. But I kept its structure the same size as another one (struct
> > > rte_eth_hash_global_conf global_conf) already in that union, they are
> > > both in 12 bytes. That means nothing changed for ' struct
> > rte_eth_hash_filter_info'.
> > 
> > First, it should be justified in the commit message.
> > Second, I think it's wrong:
> > The union has struct rte_eth_hash_global_conf = enum + uint32_t + uint32_t You
> > are adding struct rte_eth_input_set_conf = uint16_t + uint16_t + enum + enum
> > I'm not sure we can assume the sizes are the same.
> Oh, yes, the size might not be the same on all platforms.
> 
> > 
> > The ethdev ABI is already broken in this release. So maybe we can agree that it's
> > not a big deal but it should be notified in the release notes and, in this case, you
> > could keep your original patch with 128 enums.
> That's bad, but good for me. Can I just enable v6 as 'new', and disable 'v7'?
> Or I need a new version?

I think you need a new version to notify ABI change in the release notes.