Re: [PATCH v28 14/22] selftests/x86: Add a selftest for SGX

[Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>]

On Thu, Mar 05, 2020 at 01:33:28PM +0200, Jarkko Sakkinen wrote:
> On Wed, 2020-03-04 at 14:27 -0500, Nathaniel McCallum wrote:
> > > > +xsave_area:
> > > +       .fill   1, 4, 0x037F            # FCW
> > > +       .fill   5, 4, 0
> > > +       .fill   1, 4, 0x1F80            # MXCSR
> > > +       .fill   1, 4, 0xFFFF            # MXCSR_MASK
> > > +       .fill   123, 4, 0
> > > +       .fill   1, 4, 0x80000000        # XCOMP_BV[63] = 1, compaction mode
> > > +       .fill   12, 4, 0
> > 
> > I find this much more readable:
> 
> And I always aim to get things more readable. Thank you.
> 
> > xsave_area:
> >         # Legacy
> >         .fill   1, 4, 0x037F            # FCW
> >         .fill   5, 4, 0
> >         .fill   1, 4, 0x1F80            # MXCSR
> >         .fill   1, 4, 0xFFFF            # MXCSR_MASK
> >         .fill   60, 8, 0
> > 
> >         # Header
> >         .fill   1, 8, 0                 # XSTATE_BV
> >         .fill   1, 8, 1 << 63           # XCOMP_BV (compaction mode)
> >         .fill   6, 8, 0
> > 
> > Also, since people are likely to copy this code for their own
> > enclaves, it would be helpful to document which flags are set in FCW
> > and MXCSR.
> 
> It was meant as a test program but I'd guess what you say is true
> because it also might be the only alternative user space to Intel's
> :-) And a great starting point if you want to do things from scratch.
> 
> Because I meant it as a smoke test program for SGX, not everything is
> too well documented but given the multipurpose use for that code I'll
> make the improvements that you are suggesting.

For FPU Control World (FCW), I think 0x037F is not the right value even
if section 18.5 in the x86 SDM says that it is the initial value for it.

I took that value from that section.

The reason I think that there is an error in the SDM is that if you look
at the section 8.1.5, you'll see that bit 6 is a reserved bit. Thus,
does not make to set it on.

I think the legit value ought to be 0x33F i.e. unset bit 6.

In any case check:

https://raw.githubusercontent.com/jsakkine-intel/linux-sgx/master/tools/testing/selftests/x86/sgx/encl_bootstrap.S

I.e. both have now a reference:

1. To the section that describes the default.
2. To the section that describes what the bits mean.

/Jarkko