[Cbe-oss-dev] gang scheduling

[Eric Blossom]

On Wed, Feb 28, 2007 at 01:15:51AM +0100, Arnd Bergmann wrote:
> On Tuesday 27 February 2007, Christoph Hellwig wrote:
> > On Tue, Feb 27, 2007 at 01:30:23PM +0100, Arnd Bergmann wrote:
> > > a reference to the task_struct in the spu_context while we're inside
> > > of spu_run(), but then we can always access the up-to-date priority
> > > data from the right place.
> > 
> > As soon as we have our own dynamic timeslicing we will need to actually
> > change prio, and the cpu scheduler will get badly confused if you modify
> > it underneath.
> 
> Right, if you want to store dynamic priority for the context, it obviously
> needs to be stored in there. I was assuming that it is enough to store
> some information about the length and end of the previous time slice in
> there in order to determine the length of the next one, but I guess that
> is equivalent to storing a dynamic priority.
> 
> However, the static priority (nice value), rt_priority and policy should
> not be modified by the SPU scheduler AFAICS, so I see no reason to create
> redundant copies that easily get outdated.
> 
> 	Arnd <><


Where does "gang scheduling" fit in all of this?

We're looking at using direct SPE -> SPE DMA for part of our GNU Radio
port to the cell.  Is there something I have to do to ensure that when
my code is scheduled, it's really scheduled across all of it's
logically assigned SPEs?

Thanks,
Eric