On 24.11.2021 19:47, Andrew Cooper wrote: > Real hardware behaviour does not trim max leaf when certain features are > turned off, and will report blocks of trailing zeros.
I question that, but I'm also unaware of a specific case where feature disabling (presumably in the BIOS) would lead to a trailing unpopulated leaf. > None of the software manuals permit any inference based on max leaf, > which is why the 4.15 behaviour has been fine for the lifetime of Xen so > far. Yet the behavior with AMX (or KeyLocker) becomes quite odd: VMs would see lots of trailing empty leaves by default on hardware supporting these, as soon as we have (opt-in) support for them. That's because of the large gap of leaves we're not making use of just yet. The manual not permitting inference doesn't mean people can't infer things. Whether they can do any bad from this (most likely just to themselves) is unclear in the general case. To be clear - this isn't an objection to the proposed revert. But the aspect wants addressing imo, and not only in many years time, nor by simply continuing to ignore the AMX work which I did submit over half a year ago. (I didn't even dare to submit the partial KeyLocker work I've done, both for this reason and for it being just partial work.) Jan
