• State Accepted Answer
  • Locked Locked
  • Replies 4 replies
  • Subscribers 26 subscribers
  • Views 2982 views
  • Users 0 members are here
2025 survey

We are running a survey to help us improve the experience for all of our members. If you see the survey appear, please take the time to tell us about your experience if you can.

Options
Related
How was your experience today?
This discussion has been locked.
You can no longer post new replies to this discussion. If you have a question you can start a new discussion

Behaviour of cheri_pcc_get() on Linux hybrid-cap and Morello Compartments Demo

PeteD

Hi

I have been porting the Morello compartments demo to Debian linux, cross-compiling with g++ and link with LLVM lld.

I have noticed this attempts to use cheri_pcc_get() to build a capability with same bounds/permissions/tag/flag as the PCC, however in Linux hybrid-cap for me this will only read the address part.

Is this the expected behaviour of cheri_pcc_get() in hybrid?

This built-in is being compiled to ADRP <reg> by the gcc compiler.  According to the Morello architecture reference, in hybrid then ADRP will only read the offset or value which would seem to make sense.  A quick check with clang suggests the builtin is being compiled in the same way.

So, this would mean that one of the following is true:

  1. My understanding of cheri_pcc_get() is wrong and/or for some reason things are different on Linux hybrid than Android (which the demo was targetting)
  2. The gcc and clang compilers are not generating the correct assembly for the builtin on Morello
  3. The Morello compartments demo doesn't work

What say you?

Thanks
Pete
(P.S: The port was made to work by using cheri_ddc_get() since that also covers the whole memory range, although that seems a bit of a cludge)

Parents
  • 0 in reply to Jacob Bramley

    No Jacob you are right, Clang does it properly... apologies, the problem was that to check what clang produced I simply compiled the C code and then used objdump to view the assembly.  Unfortunately, as objdump not updated for Morello, the CVTP instruction was not interpreted and so from a quick look the instruction was ADR x8, <value> which I took to be the same as gcc!  Did not investigate it further at that time, sorry again for confusion!

Reply
  • 0 in reply to Jacob Bramley

    No Jacob you are right, Clang does it properly... apologies, the problem was that to check what clang produced I simply compiled the C code and then used objdump to view the assembly.  Unfortunately, as objdump not updated for Morello, the CVTP instruction was not interpreted and so from a quick look the instruction was ADR x8, <value> which I took to be the same as gcc!  Did not investigate it further at that time, sorry again for confusion!

Children
No data