Steve -

I think I am using the latest 2.0 beta (m5-2.0b2).  I downloaded it at the
beginning of April, I think.  I'm not sure that it's the bug that you
mentioned, though, because I think I was basically seeing the same
statistics if I ran for 100 mil ticks or 1 bil ticks or 10 bil, etc.  So,
I'm guessing maybe it really was in livelock...

I was (inadvertently!) running the scripts I got from Ron after my April 22
post in 03 CPU mode (even though he warned me that he wasn't sure they'd
work in 03).  When I ran in simple CPU, I didn't get the livelock messages
and my stats actually did look better.  So perhaps it was something about
the way I was running it and setting up the coherence protocols that is
actually what was causing this.  I managed to get the data I needed using
Ron's patches and scripts in simple CPU mode, so I think I made out ok.

Thanks,
-Edith

On 5/10/07, Steve Reinhardt <[EMAIL PROTECTED]> wrote:

Sorry for taking so long to reply.

You're correct that the message about repeated store-conditional
failures is supposed to be a warning about potential livelock.  The
livelock would be more at the coherence protocol level than in the
application though.

What version of m5 are you using?  I believe in the recent 2.0 beta
releases there's a bug where the failed-store-conditional counter
doesn't get reset, so it's possible to get this warning even when there
isn't a problem.

Steve

Edith Hand wrote:
> Hello again...
>
> Has anyone ever seen the subject error.  I'm in the process of using
> --trace_flags and adding DPRINTFs to try and figure out exactly what's
> going on.
>
> My interpretation of this error is that the store conditional is
> basically always failing (I repeatedly get this message in increments of
> 100000).  So, that's sounds pretty bad...  It makes me wonder if the
> code being simulated is deadlocking (if it's using LL/SC to implement
> locking and the SC keeps failing, maybe it can never acquire a lock?).
>
> If anyone has seen this type of message and know what it means, I'd
> appreciate any feedback you can provide.
>
> FYI, I'm running ALPHA_SE.
>
> Regards,
> -Edith
>
>
> ------------------------------------------------------------------------
>
> _______________________________________________
> m5-users mailing list
> [email protected]
> http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

_______________________________________________
m5-users mailing list
[email protected]
http://m5sim.org/cgi-bin/mailman/listinfo/m5-users

Reply via email to