Syberkonda Theoretically speaking, would Google be able to dedicate CPU cache to hardware acceleration via software? From my understanding, it seems like Google made a software choice to redirect CPU cache elsewhere and that this is theoretically "reversible."
I haven't yet seen a clear explanation of the change. If the change was (hypothetically) to remove L1 caching of tags, then tag checking will probably be a lot slower and that's it. If the change was (hypothetically) to shrink the amount of tag cache at each level or a couple of levels, tag checking will be slower to an extent that might be hard to estimate. If (hypothetically) tags aren't cached, tag checking would probably be genuinely quite slow.
Without knowing what was changed and what knobs there are, it's hard to guess what might happen if the knobs were twisted.
Of course, Google could at any point say what was changed.
I'm pretty confident that the GrapheneOS developers will provide more information if/when they know more. I also suspect they're not going to say a lot more before they know more.
Some things that may well be up in the air:
- Given Google's current firmware, how much slowdown would MTE incur for plausible workloads?
- Is it possible to use Google's current firmware to gather data to guess at what changes were made?
- With Google's current firmware, does MTE work reliably? That is, are there any problems aside from performance?
That sort of investigation could take a day or two, or a week or two, or a month or two. If Google releases another firmware update, that could reset the clock.