[openfabrics-ewg] OFED-1.1-rc4 is ready

[Doug Ledford]

On Wed, 2006-09-13 at 09:25 +0300, Michael S. Tsirkin wrote:
> Quoting r. Bryan O'Sullivan <bos at pathscale.com>:
> > > the ibv_driver_init function was changed to openib_driver_init. 
> >
> > By the way, I find it unsettling that the current libibverbs internal
> > ABI allows silent breakage like this that cannot be detected except at
> > runtime, and then only when the right hardware is present.
> > 
> > Mind you, I don't have any better suggestions in mind (at least not at
> > 10:30pm).
> > 
> > But I worry about the possibility this leaves open for botched field
> > upgrades breaking userspace in you-don't-find-out-until-it's-too-late
> > ways when libibverbs 1.1 starts being used.
> 
> libipathverbs can simply export both ibv_driver_init and
> openib_driver_init like libmthca does, that's what we'll do for OFED.
> 
> Or maybe Doug here can come up with some symbol versioning trick.
> Dough?

I don't think you can do symbol versioning here.  For symbol versioning
to work you have to have a compile time map from the source used to the
version you are linking to.  For all the drivers, like mthca, they are
compiled after libibverbs, and so libibverbs is built blind to the
drivers if you will, yet it is the drivers that provide the symbol and
therefore the symbol version according to the linker, so libibverbs can
never have the automated type symbol versioning.


-- 
Doug Ledford <dledford at redhat.com>
              GPG KeyID: CFBFF194
              http://people.redhat.com/dledford

Infiniband specific RPMs available at
              http://people.redhat.com/dledford/Infiniband
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://lists.openfabrics.org/pipermail/ewg/attachments/20060913/d7ac0507/attachment.sig>