I can't point to exactly what doesn't work, but software that used to run no longer runs with this CL. Another user reported similar issues, although I don't know what they were running. Essentially we're telling the simulated software it can use features that aren't actually implemented (or don't make sense, like SVM on an intel system). If the host CPU doesn't get upset, the simulated software might try actually using those features, and then something bad will likely happen.

Having comments are a good thing, but those comments weren't great. Those CPUID results hold more than just feature bits, first of all. Second, it's not well defined to say, effectively, that everything is enabled except such and such. The set those exceptions are from will change over time, making the comment invalid. It's also more relevant what features enabled than which are disabled.

My intention is to revert the original change here, and then to put up a new version which does what the original was trying to do. I have that CL posted over here http://reviews.gem5.org/r/2591/ which you reviewed. The features that needed to be enabled to get things working for KVM in SE mode (why those I have no idea) were a small subset of those enabled by this change. That was a while ago, but I think I also turned on one or two other things which we support but weren't telling software about.

Ah, I see... I had the order of these two patches reversed in my head, hence my confusion.  Makes a lot more sense now.  I'll just say that adding some comments to 2591 similar to the ones you're getting rid of here would be nice.