Re: [PATCH v11 13/20] target/riscv: mmu changes for zicfiss shadow stack protection

[Deepak Gupta <debug@rivosinc.com>]

On Thu, Aug 29, 2024 at 10:03:04AM +1000, Alistair Francis wrote:
>On Thu, Aug 29, 2024 at 9:45 AM Deepak Gupta <debug@rivosinc.com> wrote:
>>
>> On Thu, Aug 29, 2024 at 09:29:49AM +1000, Alistair Francis wrote:
>> >On Thu, Aug 29, 2024 at 3:49 AM Deepak Gupta <debug@rivosinc.com> wrote:
>> >>
>> >>          prot |= PAGE_WRITE;
>> >>      }
>> >> -    if (pte & PTE_X) {
>> >> +    if (rwx & PTE_X) {
>> >>          bool mxr = false;
>> >>
>> >>          /*
>> >> @@ -1160,7 +1177,7 @@ restart:
>> >>
>> >>      if (!((prot >> access_type) & 1)) {
>> >>          /* Access check failed */
>> >> -        return TRANSLATE_FAIL;
>> >> +        return sstack_page ? TRANSLATE_PMP_FAIL : TRANSLATE_FAIL;
>> >
>> >Why is it a PMP error if it's a shadow stack page?
>>
>> A shadow stack page is readable by regular loads.
>> We are making sure of that in `case PTE_W` in above switch case.
>> But shadow stack page is not writeable via regular stores. And must raise
>> access fault. return code `TRANSLATE_PMP_FAIL` is translated to access fault
>> while raising fault.
>
>Ah, ok. It's worth commenting that we are returning TRANSLATE_PMP_FAIL
>as that will be translated to an access fault

Ack.

>
>Alistair
>
>>
>> >
>> >Alistair