Murali Vijayaraghavan comments

Results 51 comments of


                                            Murali Vijayaraghavan

Simplifying WARL address CSR writes

Makes sense!

Taken branches and Jumps checking if the taken address is less than `topBound-2` is not useful

> This has been a reason for long discussions both. You will always need the bounds check in instruction fetch. Every instruction changes the PCC and thus can fall out...

Taken branches and Jumps checking if the taken address is less than `topBound-2` is not useful

@tariqkurd-repo

Taken branches and Jumps checking if the taken address is less than `topBound-2` is not useful

Thanks @arichardson. I skimmed through base RISC-V unprivileged and privileged specs. Bounds check is brand-new behavior in CHERI. In RISC-V, one could potentially have a page fault on a 4-byte...

Taken branches and Jumps checking if the taken address is less than `topBound-2` is not useful

> However, without a way to determine this, you could create an OOB capability to another compartment and when invoked it would appear that the target compartment did somethign illegal...

Taken branches and Jumps checking if the taken address is less than `topBound-2` is not useful

> This would leave the architecture unchanged, but eliminate the duplicate bounds check? It doesn't need a new CSR, but the semantics of `MEPCC` changes. It does avoid the duplicate...

Taken branches and Jumps checking if the taken address is less than `topBound-2` is not useful

Yeah, you don't need another functional unit, but the inputs of that functional unit now has to deal with JALR/JAL also; and the output has to go into the priority...

Taken branches and Jumps checking if the taken address is less than `topBound-2` is not useful

This is a valid point! We now have to do representability checks in Jumps/taken branches. However, the higher level issue remains: we cannot determine if the target address contains a...

Taken branches and Jumps checking if the taken address is less than `topBound-2` is not useful

> Sure it is. Don't give out those capabilities if you don't want people to use them. The scenario I mentioned shows why it's not always the case that one...

Taken branches and Jumps checking if the taken address is less than `topBound-2` is not useful

> I'm strongly opposed to adding new architectural state to debug control transfers when we have perfectly good E-Trace and CTR extensions that should compose with CHERI. We don't need...