On 07/15/2013 04:05 PM, Nilay Vaish wrote:
On Mon, 15 Jul 2013, Ali Saidi wrote:
On 15.07.2013 12:44, Nilay Vaish wrote:
We are not talking about
the fetchStatus here.
--
Nilay
Well your change touches both. I
don't see how fetch status becoming idle can ever occur except for the
change you have.
Let me repeat, I was never ever talking about fetchStatus. If you want
to talk about the proposed to change to fetchStatus, explicitly say
that. How can you take my remarks in context of cpu state being idle and
start applying them in the context of fetchStatus being idle?
ARM doesn't micro-code the quiesce operation that calls
suspend() so I don't really care (if we did this wouldn't work for ARM),
but I still think Andreas Sandberg's point about the definition of the
instruction in the manual saying that it doesn't always jump to the
interrupt vector is a concern. Either way I think it's Andreas' or
someone from AMD that probably cares the most. If they're fine with it
then I am too.
You should read the entire thread. If my understanding of the response
from Andreas (date 28th June, 2013) is correct, then he conceded that
the proposed change is correct.
Yes, the change is /probably/ correct. However, I'm opposed to the
change because it makes draining needlessly complicated. I'm still of
the opinion that this is an x86 frontend issue and that should be fixed
in the ISA code (e.g. by adding a hlt+eret micro-op) rather than
creating a mess in the backend.
We really need to define the CPU/thread states at some point and
document when and why transitions occur to prevent this kind of
confusion in the future, but that's a separate issue.
//Andreas
_______________________________________________
gem5-dev mailing list
[email protected]
http://m5sim.org/mailman/listinfo/gem5-dev
