[PATCH 2/2] confdefs.h hack

[Joel Sherrill]

On Tue, Mar 29, 2016 at 7:48 AM, Gedare Bloom <gedare at rtems.org> wrote:

> On Tue, Mar 29, 2016 at 7:51 AM, Sebastian Huber
> <sebastian.huber at embedded-brains.de> wrote:
> > On 29/03/16 13:47, Sebastian Huber wrote:
> >>
> >> Not to be committed. How can we fix this?
> >
> >
> > I was able to run all MP tests except mp14 successfully using Qemu and a
> > virtual network. The TCP/IP based MPCI support is basically BSP
> independent
> > and should work for all BSPs that support a network interface driver in
> the
> > old network stack. So, maybe this MPCI driver should move to libchip. I
> am
> > not sure how we should address the confdefs.h issue.
> >
> This makes sense. You could add a new configure flag for the driver
> then, e.g. CONFIGURE_MPCI_DRIVER_ENABLED or something to control the
> confdefs ++bloat.
>
>
There is already CONFIGURE_MP_MPCI_TABLE_POINTER which is really
an older name for what you are suggesting. For consistency, it could be
renamed.

You should have "mpci net" and "mpci shm" and let the bsp.h select. There is
precedence for bsp.h to set defaults for confdefs.h.

Disclaimer: And yes.. the public interface for bsp.h isn't as well defined
as we
would like but I am trying to trim down what's included and start moving
things
to bspimpl.h to get a few examples that are cleaner.

Wasn't your other configuration issue that you wanted a server thread and
some resources per core?


> > The mp14 test seems to use a global partition. Does this make sense in a
> > non-shared memory setup, e.g. message passing via TCP/IP?
> >
> I doubt it makes sense. You need to have a lot of middleware to get a
> (partitioned) global address space to work over a message passing
> interface. It has been done before e.g. to use UPC/OpenMP over MPI,
> but I don't think this is necessarily good in our case.
>
>
No. It does not make sense.

Long ago, there was an MPCI implementation over MPI but I personally
never saw it. I don't know how they dealt with this.

Thinking generically, I would say we need a way to say global partitions
are not
supported by a particular MPCI.

FWIW should the "HAS OWN TABLE" support in confdefs.h disappear? I would
really hope that no one ever writes their own configuration by hand. :)


> > --
> > Sebastian Huber, embedded brains GmbH
> >
> > Address : Dornierstr. 4, D-82178 Puchheim, Germany
> > Phone   : +49 89 189 47 41-16
> > Fax     : +49 89 189 47 41-09
> > E-Mail  : sebastian.huber at embedded-brains.de
> > PGP     : Public key available on request.
> >
> > Diese Nachricht ist keine geschäftliche Mitteilung im Sinne des EHUG.
> >
> >
> > _______________________________________________
> > devel mailing list
> > devel at rtems.org
> > http://lists.rtems.org/mailman/listinfo/devel
> _______________________________________________
> devel mailing list
> devel at rtems.org
> http://lists.rtems.org/mailman/listinfo/devel
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.rtems.org/pipermail/devel/attachments/20160329/38744cb2/attachment-0002.html>