MemHierarchy should not segfault on misaligned loads

[ghost]

L1CoherenceController can segfault when the cache receives a request that spans across cache lines. It would be nicer to Output.fatal() with such a message, rather than segfault.

[nmhamster]

@bjmoor @gvoskuilen - I think this depends on the contract with the processor core people. If the requirement is that the core ensures no unaligned/split requests then its OK to not have this checked. It might also be better to not have the checks slowing down the behavior? Thoughts?

[gvoskuilen]

@nmhamster @bjmoor - I think it should be a requirement for the core to ensure no unaligned/split requests. I'd rather avoid an extra check on every access.

[ghost]

Personally, I think that branch predictors are pretty good, and segfaults are bad. Users should not get a segfault using our code. That puts a significant onus on them how to debug code that isn't theirs, and when I was receiving that segfault, it was FAR from obvious that the reason was mis-aligned cache accesses.

[allevin]

Fixed if user turns on debug.