Nice!  This is actually way simpler than I expected.

Seems like at some point we'd want to (1) set the simulation quantum from the main script based on the etherlink delay and (2) maybe have a check in the etherlink object that asserts that if the SimObjects on the two ends belong to different event queues, then the simulation quantum is at least as large as the latency.  I guess #2 is optional since you'd get an assertion once the simulation started running and handleAsyncInsertions() tried to schedule an event in the past (that's what should happen, anyway), but it would be friendlier to generate a cleaner error message up front.

configs/common/FSConfig.py (Diff revision 1)

It looks like this wouldn't work for the realview system, is that right?

Show all issues

1. Nilay Vaish Feb. 10, 2013, 9:06 p.m. (Feb. 10, 2013, 9:06 p.m.)

   Yes. As of now I have only tested with the Alpha regression test.
   Arm would require its own set of changes, though it might be that
   a couple of lines in this script and only the decoder needs to change.

tests/configs/twosys-tsunami-simple-atomic.py (Diff revision 1)

I think this is redundant, right?  drive_sys should inherit eventq_index from root.  For that matter, root should default to 0 as well, so only the 'test_sys.eventq_index = 1' line is strictly necessary.  I'm not opposed to leaving the other lines in for documentation purposes.

1. Nilay Vaish Feb. 10, 2013, 9:06 p.m. (Feb. 10, 2013, 9:06 p.m.)

   Yes, you are right.

Like Steve i find the changes required to be remarkably small (which is great!). Seeing as there is some subtleness to what is going on it would be very nice to have some comments about what is going on here. Otherwise I'm happy.

Thanks again for making this happen,
Ali

src/dev/Ethernet.py (Diff revision 1)

A comment about what there are would be great?

Show all issues

src/dev/etherlink.hh (Diff revision 1)

same here please

Show all issues

configs/common/FSConfig.py (Diff revision 1)

It's a separate issue, but somehow it would be good to solve the naming or have some kind of alias for the common bits that does not involve tsunami, realview, pc etc. That would make the scripts much neater.

Show all issues

src/dev/Ethernet.py (Diff revision 1)

Is it really SimObject we need? Is it not EventManagers?

Perhaps I'm missing the point.

Show all issues

1. Nilay Vaish Feb. 16, 2013, 12:29 a.m. (Feb. 16, 2013, 12:29 a.m.)

   You right that an event manager should be enough. But, event managers
   cannot be passed as params. And I don't see the need to type cast
   these sim objects to event managers.

src/dev/etherlink.cc (Diff revision 1)

I would be good to add a check that the values are sensible (no NULL), unless that's covered by the Python machinery.

Show all issues

1. Nilay Vaish Feb. 16, 2013, 12:29 a.m. (Feb. 16, 2013, 12:29 a.m.)

   Since I am not assigning any default value to these params, IIRC, some
   python script does complain that the params do not have any assigned / default
   value.

configs/common/FSConfig.py (Diff revision 1)

Is there a way we can derive int0/int1 from end0/end1, or vice versa?  It seems redundant (and potentially error-prone) to have to set them separately.

Show all issues

1. Nilay Vaish Feb. 16, 2013, 1:45 a.m. (Feb. 16, 2013, 1:45 a.m.)

   While it should be possible to do so, it does not seem straight forward enough to me.

src/dev/etherlink.cc (Diff revision 1)

Does it make sense that we're only serializing one end of the link here?

Show all issues

1. Nilay Vaish Feb. 16, 2013, 1:45 a.m. (Feb. 16, 2013, 1:45 a.m.)

   I think that change is not correct. I rolling back some of the proposed changes
   so as to take care of this.