On Tue, 16 Jun 2020 10:45:49 +0200
Thomas Monjalon <tho...@monjalon.net> wrote:
> 16/06/2020 10:33, Tal Shnaiderman:
> > > 01/06/2020 11:38, tal...@mellanox.com:
> > > > From: Tal Shnaiderman <tal...@mellanox.com>
> > > >
> > > > Export needed eal functions used by the lib.
> > > >
> > > > Signed-off-by: Tal Shnaiderman <tal...@mellanox.com>
> > > > ---
> > > > --- a/lib/librte_eal/rte_eal_exports.def
> > > > +++ b/lib/librte_eal/rte_eal_exports.def
> > > > @@ -3,6 +3,7 @@ EXPORTS
> > > > per_lcore__rte_errno
> > > > rte_calloc
> > > > rte_calloc_socket
> > > > + per_lcore__lcore_id
> > > [...]
> > > > --- a/lib/librte_eal/rte_eal_version.map
> > > > +++ b/lib/librte_eal/rte_eal_version.map
> > > > @@ -394,4 +394,5 @@ EXPERIMENTAL {
> > > > rte_mem_unmap;
> > > > __emutls_v.per_lcore__rte_errno;
> > > > __rte_trace_mem_per_thread_alloc;
> > > > + __emutls_v.per_lcore__lcore_id;
>
> Note the alphabetical order is not respected above.
>
> > > I'm surprised about this one in the Linux/BSD map.
> > > Can it be removed?
> >
> > It is needed for the MingGW build which requires the __emutls_v. prefix.
>
> We already have per_lcore__lcore_id and per_lcore__rte_errno
> for Linux and FreeBSD.
> The prefix __emutls_v. for MinGW looks wrong.
> If it is really required, it should be added automatically
> in my opinion.
> I think we can do it in buildtools/map_to_def.py
> by adding this prefix to per_lcore__ when required.
MinGW doesn't use .def, it's for link.exe only (used by Clang). To my
understanding, multiple .map files can be used [1]. The "overlay .map" for
MinGW can then be generated from the generic .map by a separate script.
[1]: https://github.com/PlushBeaver/dpdk/blob/windows/lib/meson.build#L174
--
Dmitry Kozlyuk
