Re: [GIT PULL] x86/shstk for 6.4

[Mark Brown]

On Mon, May 15, 2023 at 02:22:55PM -0700, Deepak Gupta wrote:
> On Sun, May 07, 2023 at 04:24:24PM +0000, Edgecombe, Rick P wrote:

> > BTW, I forgot to mention that there is another architecture (maybe 2)
> > that is expected to use this refactor for implementing their shadow
> > stacks. So FWIW, this churn is not just for x86.

> That's right, one of them is RISC-V.

Also arm64.

> RISC-V control-flow integrity: https://github.com/riscv/riscv-cfi

> Since RISC-V PTE have 3 separate bits for read, write and execute. Write
> only (R=0, W=1, X=0) encodings had been reserved and thus cpu supporting
> this extension will treat this reserved encoding as shadow stack.

> It doesn't get messy as in case of x86 (due to overloading of dirty bit),
> but it still will need pte helper which marks a page "shadow stack
> writeable" or "regular writeable" depending on vma.

For arm64 GCS (our shadow stack equivalent) is built on top of another
extension that allows us to assign arbitrary meanings to four of the
bits (they become an index into an array of actual permissions) so we
might be able to avoid having to look at the VMA, though we might want
to in future in order to make better use of the other features of the
indirection extension.

Attachment: signature.asc
Description: PGP signature