On Thu, Jul 09, 2015 at 05:51:16PM +0200, Thomas Monjalon wrote:
> 2015-07-08 14:10, Bruce Richardson:
> > On Mon, Jul 06, 2015 at 03:16:01PM +0200, Thomas Monjalon wrote:
> > > 2015-07-02 16:16, John McNamara:
> > > > --- a/doc/guides/rel_notes/abi.rst
> > > > +++ b/doc/guides/rel_notes/abi.rst
> > > > Deprecation Notices
> > > > -------------------
> > > > +
> > > > +* In DPDK 2.1 the IEEE1588/802.1AS support in the i40e driver makes
> > > > use of the
> > > > + ``udata64`` field in the mbuf to pass the timesync register index to
> > > > the
> > > > + user. In DPDK 2.2 this will be moved to a new field in the mbuf.
> > >
> > > We need more acknowledgements for this decision, as stated here:
> > > http://dpdk.org/browse/dpdk/tree/doc/guides/guidelines/versioning.rst#n51
> >
> > Why can't this new field just be added at the end of cache line 1 (the
> > second
> > cache line) of the mbuf? That would avoid any ABI breakage and would mean we
> > can just put the change in in this release, instead of waiting.
>
> Are you sure that (because of __rte_cache_aligned) the size of the structure
> is never increased with this new field?
> Please confirm your opinion.
This is checked at compile time by the test app.
930 static int
931 test_mbuf(void)
932 {
933 RTE_BUILD_BUG_ON(sizeof(struct rte_mbuf) != RTE_CACHE_LINE_SIZE *
2);
....
So if a change does result in an increase the mbuf size, by causing overflow in
either the first or the second cache line, we will get compiler errors in the
build because of it. Therefore, such changes are pretty easy to test by
compiling
up on our supported targets.
/Bruce
