Solved bugs while switching cpu models with the --repeat-switch command

Information
Submitter:	Alberto Javier Naranjo Carmona
Repository:	gem5
Branch:
Bugs:
Depends On:
Reviewers
Groups:	Default
People:

Description

This patch solves some bugs that occured when using the option --repeat-switch, in particular when switching from O3 to timing model (deadlocks while draining and assert(!memReq) in drainSanityCheck()). See the following thread for more info http://www.mail-archive.com/gem5-users%40gem5.org/msg10631.html

Testing Done

All long regression tests passed. tsunami-switcheroo-full passed.

Issue Summary

Description	From	Last Updated	Status
is this change supposed to be in here?	Ali Saidi	Nov. 7, 2014, 1:39 p.m.	Resolved
I'm not sure which version of gem5 you're using, but in the latest dev repository it looks like this if ...	Ali Saidi	Nov. 7, 2014, 1:39 p.m.	Resolved

Diff:

Revision 2 (+9 -6)

Show changes

	src/cpu/o3/fetch_impl.hh
	src/cpu/o3/lsq_impl.hh
	src/cpu/o3/lsq_unit.hh
	src/mem/ruby/system/RubyMemoryControl.cc

thanks for the patch. it looks good, but I'm not sure if the ruby change it supposed to be in there. otherwise I think it's good to go.

src/mem/ruby/system/RubyMemoryControl.cc (Diff revision 2)

is this change supposed to be in here?

Show all issues

Alberto Javier Naranjo Carmona Oct. 27, 2014, 10:02 a.m. (Oct. 27, 2014, 10:02 a.m.)

Yes, I experienced deadlocks in this line while draining. Suppose a store request missed in the cache and was scheduled in the memory controller. Then, the drain signal gets trigeered and the memory controller deschedules all the events (including the store request). However, the LSQ will never get drained because is waiting for the memory controller to answer a request that was just discarded.

src/cpu/o3/lsq_impl.hh (Diff revision 2)

I'm not sure which version of gem5 you're using, but in the latest dev repository it looks like this if block is completely gone. I'm not sure if that is because the check has been moved elsewhere or otherwise.

Show all issues

Alberto Javier Naranjo Carmona Oct. 27, 2014, 10:06 a.m. (Oct. 27, 2014, 10:06 a.m.)

Actually I found this bug using the latest stable repository. I wasn't aware of these changes in the dev repository. May be this change doesn't apply anymore

src/cpu/o3/lsq_impl.hh (Diff revision 2)

If you have a chance to test the dev repository that would be great.

Alberto Javier Naranjo Carmona Nov. 7, 2014, 1:36 p.m. (Nov. 7, 2014, 1:36 p.m.)

I was busy but I tested the dev repository, it seems like the changes in the files src/cpu/o3/lsq_unit.hh and src/cpu/o3/fetch_impl.hh are not needed anymore. Also the file src/mem/ruby/system/RubyMemoryControl.cc was moved to src/mem/ruby/structures/RubyMemoryControl.cc
I will update the diffs

Change Summary:

The previous diff was done with the stable repository and not all the changes matched with the current dev repository.
Did the same changes (at least those that still apply) and run the same tests as before. Everything seems OK

Diff:

Revision 3 (+1 -4)

Show changes

	src/cpu/o3/fetch_impl.hh
	src/mem/ruby/structures/RubyMemoryControl.cc

the o3 change seems fine. . someone more familiar with ruby would need to comment on the other one.

You have a pending review.

Review Board 2.0.15

This change has been marked as submitted.

Screenshots

Files

Issue Summary

Diff:

Change Summary:

Diff:

Status: Closed (submitted)