[Cbe-oss-dev] [PATCH 6:9] Implement deferred queuing for spu class 1 faults

[Kazunori Asayama]

Luke Browning wrote:
> Implement deferred queuing for spu class 1 faults
> 
> Yield the spu when processing class 1 faults as they take a 
> long time to be resolved.  Activate after the fault has been 
> resolved so that the context may be scheduled.
> 
> Time slice contexts in this state but don't put them on the
> run queue.  Let the fault handling code re-queue the context
> when it is ready to run.  
> 
> In general, we don't want spus in the faulting state to be 
> added to the run queue as it gives a false indication that 
> there are jobs waiting to run which may cause a bunch of 
> secondary context switches, particularly, in the time slicing 
> code.  Nor do we want to resume a context in this state if 
> there are other running contexts to be resumed.  Therefore,
> we need to keep contexts that are processing class 1 faults
> off the run queue.  We don't really care about the other
> types of faults because they are resolved quickly.
> 
> Signed-off-by: Luke Browning <lukebrowning at us.ibm.com>
> 
> ---
> This fixes the Cntrl-C hang.  This patch is intended to 
> replace the previous version of this patch.
> 
> ---

It works.

Thank you,
-- 
(ASAYAMA Kazunori
  (asayama at sm.sony.co.jp))
t