[openib-general] Re: [PATCH] rdma_lat-09 and results

[Michael S. Tsirkin]

Quoting r. Roland Dreier <roland at topspin.com>:
> Subject: Re: [PATCH] rdma_lat-09 and results
> 
>     Michael> At least for this usage, I think it would be best to
>     Michael> rename mthca.a (and mthca.so) to libmthca.a (and
>     Michael> libmthca.so), respectively, and put them in libpath, then
>     Michael> softlink libmthca.so to infiniband directory.
> 
>     Michael> This avoids the need for such utilities.  Would you
>     Michael> accept such a change?
> 
> That makes sense, although I don't see why we want libmthca.so in the
> regular library directory.

I can imagine a cluster system where loading shared libraries work, but
reading a directory doesnt.
Ronald Minnich should be able to enlighten us on whether bproc really
has such a limitation. Roland - do I describe brpoc correctly?
I know bproc supports dlopen from remote applications, but
what about glob() on a remote directory?  Does that work from a blade?

> It seems like the best would be to have
> libmthca.a in $(libdir) and mthca.so in $(libdir)/infiniband.  That
> way we make it easy to do static linking and avoid having two names
> for the same object.

OK, but I dont know how to do that in autotools. Do you? 

>     Michael> Hmm, this would be a security problem if someone sets
>     Michael> suid bit on programs using uverbs.
> 
> True, good point.  To me being able to set the search path is much
> more useful than being able to have secure suid verbs programs,

Actually, whats the importance of it? It seems the user can compile
libibverbs as readily as the plugin ...

> so I
> don't want to get rid of the OPENIB_DRIVER_PATH.
> Does something like the following seem good enough?

Solves this particular problem, but maybe, additionally, there should be a
configuration option to disable OPENIB_DRIVER_PATH lookup?

I also wander about the user of strdupa - can it cause stack overflow?

-- 
MST