Re: [PATCH v28 21/22] x86/vdso: Implement a vDSO for Intel SGX enclave call

[Sean Christopherson <sean.j.christopherson@intel.com>]

On Fri, Mar 13, 2020 at 02:32:29PM -0400, Nathaniel McCallum wrote:
> On Fri, Mar 13, 2020 at 12:46 PM Sean Christopherson
> <sean.j.christopherson@intel.com> wrote:
> >
> > On Fri, Mar 13, 2020 at 11:48:54AM -0400, Nathaniel McCallum wrote:
> > > Therefore, I'd like to propose that __vdso_sgx_enter_enclave():
> > >   * Preserve %rbx.
> >
> > At first glance, that looks sane.  Being able to call __vdso... from C
> > would certainly be nice.
> 
> Agreed. I think ergonomically we want __vdso...() to be called from C
> and the handler to be implemented in asm (optionally); without
> breaking the ability to call __vdso..() from asm in special cases.
> 
> I think all ergonomic issues get solved by the following:
>    * Pass a void * into the handler from C through __vdso...().
>    * Allow the handler to pop parameters off of the output stack without hacks.
> 
> This allows the handler to pop extra arguments off the stack and write
> them into the memory at the void *. Then the handler can be very small
> and pass logic back to the caller of __vdso...().
> 
> Here's what this all means for the enclave. For maximum usability, the
> enclave should preserve all callee-saved registers (except %rbx, which
> is preserved by __vdso..()). For each ABI rule that the enclave
> breaks, you need logic in a handler to fix it. So if you push return
> params on the stack, the handler needs to undo that.

Or the untrusted runtime needs to wrap the __vdso() to save state that is
clobbered by the enclave.  Just want to make it crystal clear that using a
handler is only required for stack shenanigans.

> This doesn't compromise the ability to treat __vsdo...() like ENCLU if
> you need the full power. But it does make it significantly easier to
> consume when you don't have special needs. So as I see it, __vdso...()
> should:
> 
> 1. preserve %rbx
> 2. take leaf in %rcx
> 3. gain a void* stack param which is passed to the handler

Unless I'm misunderstanding the request, this already exists.  %rsp at the
time of EEXIT is passed to the handler.

> 4. sub/add to %rsp rather than save/restore

Can you elaborate on why you want to sub/add to %rsp instead of having the
enclave unwind the stack?  Preserving %rsp across EEXIT/ERESUME seems more
in line with function call semantics, which I assume is desirable?  E.g.

  push param3
  push param2
  push param1

  enclu[EEXIT]

  add $0x18, %rsp

> That would make this a very usable and fast interface without
> sacrificing any of its current power.