cache invalidation ops

[fmayer]

we should make sure that cache invalidation ops also invalidate the caches for the tags of the cacheline, which may or may not be colocated with the cachelines, depending on implementation

Adam 'pi3' Zabrocki (@Adam-pi3)

[deepak0414]

Question: software can't do it itself?

[fmayer]

You mean manually issue two cache maintenance operations? I don't think so. That would mean we have to recompile software to do that. We want heap to he tagable just by swapping out the allocator

[deepak0414]

We want heap to he tagable just by swapping out the allocator

Make sense. 👍

[lfiolhais]

My 2cs here is that the MT shouldn't define extra CMO instructions. I think it
would be easier for both hardware and software vendors to augment the behavior
of the current CMOs. For example, if a CMO instruction targets a cache block
which contains tagged data, the implementation must also flush the
corresponding tag addresses.

[fmayer]

Agreed. But we need to define this in the spec.