[Cbe-oss-dev] spu question

[Luke Browning]

Arnd Bergmann <arnd.bergmann at de.ibm.com> wrote on 10/10/2006 10:54:15 AM:

> On Tuesday 10 October 2006 15:44, Arnd Bergmann wrote:
> > > 
> > > agreed. Are you recommending modifying spe_getgroup_max() <spelling> 

> > > in the library to return the number of 'available' spus instead of 
> > > the current hardcoded value (16).
> > 
> > I think we finally did that in a recent libspe change. At least, it
> > has been planned to be done for a long time. Since you bring it up
> > again, I noticed that we may have a problem
> 
> Hmm, should have finished that sentence before hitting the send 
button...
> 
> I noticed we may have a problem on the hypervisor, because it does
> not necessarily have the SPUs in the device tree, so we can't easily
> put them into /sys/devices/system/spus either.
> 
> I wonder what we can do about this. Can we probe anything about what
> SPUs are available to Linux without other than attempting to use them?
> Or can we simply assume we have a fixed number of them by faking them
> in the flattened device tree?
> 

I don't know the dynamics of the Cell hypervisor support, but there are 
similar properties in the device tree for other power systems that 
describe max and current number of processors.  Maybe, something similar
would work here. Is there some documentation on the hypervisor model 
that you are discussing?

Luke
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.ozlabs.org/pipermail/cbe-oss-dev/attachments/20061010/2aefabe8/attachment.htm>