Closed jamie-melling closed 3 weeks ago
It looks like these rules need a thorough review. In https://github.com/riscv/riscv-cheri/pull/421 we default to RC permission where there is a choice, so maybe we should do the same here.
Typo fix here: #429
If you have a quadrant 1 capability with perms[2:0] equal to either 0-1 or 4-5 (with w-perm) and remove just
EL
, there are two possible outcomes, neither of which break monotonicity. Either you stay in quadrant 1 and stripC
andLM
to get anR
,W
andX
capability. The other option is to create a quadrant 2 capability where you stripLM
,W
andX
to get anR
andC
capability. The rules laid out in ACPERM do not cover which mutation is desired.One other note, small spec bug in ACPERM rules.