On 5/18/26 17:18, Michael Clark wrote:
rather lengthy update on a high-level architecture exploration:
- https://metaparadigm.com/~mclark/glyph.pdf
I have reached an inflection point with nested virtualization and SLAT.
I believe I expose enough from an architectural perspective to support
SLAT in this version of the architecture, theoretically, as we have a
dominating machine capability for each virtual domain. it still needs
testing in a full system architectural simulator. I think this version
of the specification is worth a fine look, though. I just made a change
to domain context switching for nested traps, keeping in mind SLAT plus
TLB miss handlers for first level address translation.
received some feedback and made several more simplifications:
- page color caps are trimmed to the subset mapping to Intel PKRU.
- the array extension (image load store) has been commented out.
- domain capability trap routing has been further simplified.
curious on feedback for the domain capabilities privilege model.
it still needs to be tested in simulation. they should be amenable
to accelerated translation as they only need to be checked on
interrupts and access to privileged registers or instructions.
user-mode plus single supervisor would occupy two domain slots.
page table colors has a pretty direct mapping to Intel PKRU.
4K pages using the default scheme without TLB miss handlers should
be amenable to fast shadow paging on Intel X86. there are still lots
of missing details. VM and instruction fences are not yet defined.
of course we need to add the instructions. that will be done last:
- https://github.com/larkmjc/x86
