PMP (Physical Memory Protection) on BeagleV Ahead

BeagleV
,
1

Hey,

I’m currently working on a research project on RISC-V where I try to protect memory regions from either OpenSBI or the kernel. Unfortunately I can’t get PMP to work on the BeagleV Ahead.

Every time I boot up the system, the OpenSBI PMP detection prints:

Boot HART PMP Count       : 0
Boot HART PMP Granularity : 0
Boot HART PMP Address Bits: 0

This detection simply checks if the PMP CSRs are writable, so I’m pretty confident that it should print 8 or 16 if the chip behaves correctly (The official C910 documentation states that there are options with either 8 or 16 PMP registers, so I assume my machine has at least 8).

My current guess is that either PMP is broken on the C910 (TH1520) or that some software I currently have no control of already sets the PMP registers and sets them locked so that I can’t change them later. What stands against this guess is that reading from the pmp registers returns 0, which I would not assume even in the locked case.

I already tried to patch the two projects I’m currently aware of that could setup PMP related functionality:

Here are my main questions:

  1. Has someone found PMP working on the C910 (TH1520)?
  2. I found that the Linux kernel sets up PMP in early boot when it is compiled with M_MODE. Is PMP required to be setup to do memory writes? I assumed when PMP is not setup it just performs no checks.
  3. Is there some other software involved in setting up PMP before U-Boot or in between the jump from U-Boot to OpenSBI?
2

(Can’t edit for some reason) Related post for LicheePi4A: PMP (Physical Memory Protection) on LicheePi4A · Issue #29 · sipeed/LicheePi4A · GitHub

3

I have the same problem with BeagleV Ahead, but unfortunately no answers. I tried to get Keystone Enclave framework (https://keystone-enclave.org/) to be ported to BeagleV Ahead. I used OpenSBI v1.4 as it seems to include support for the board as part of its generic platform. Keystone is using PMP to manage enclaves. I was hoping that it will work with small modifications despite those odd Boot HART PMP log entries but no luck so far.

4

Cross posting from the github issue again. The current state there is that I got opensbi v1.4 to work, but it didn’t help. At this point I’m pretty sure that SBI can’t be the problem.

With a few hints I found that the pmp registers seem to be mapped to memory in sbi, where sbi then clears the current configuration. But when I try writing meaningful configurations into that memory region, the writes to the interesting bits are silently dropped.