[Cbe-oss-dev] spu signals and context switches

[André Detsch]

Håvard Espeland wrote:
> I'm playing around with the spu preemptive scheduling mechanism, and
> I noticed that signals sent from the ppu to spus when I have more spu
> contexts than physical spus are sometimes lost. Is this intended, or is
> my code just broken?

Yes there is a bug when a signal is sent to a context which is not bound 
to a SPU. I.e., when the backing operation is used instead of the direct 
hardware operation.

I'm trying to figure out exactly what is happening, but, observing the 
SPEChannel information provided by the simulator, it looks like the 
problem is related to the events acks management: we are not 
saving/restoring this information on the context switchs, ending up with 
a situation where there is a ack set for a not-yet-processed event (like 
MFC_SPU_MAILBOX_WRITTEN_EVENT).

--
André Detsch