Re: [PATCH v2] SEV: QMP support for Inject-Launch-Secret

[James Bottomley <jejb@linux.ibm.com>]

On Fri, 2020-07-03 at 17:02 +0100, Dr. David Alan Gilbert wrote:
> * James Bottomley (jejb@linux.ibm.com) wrote:
> > On Fri, 2020-07-03 at 12:11 +0100, Dr. David Alan Gilbert wrote:
> > > * Tobin Feldman-Fitzthum (tobin@linux.vnet.ibm.com) wrote:
> > 
> > [...]
> > > > +    input.trans_uaddr = (uint64_t)data;
> > > > +    input.trans_len = data_sz;
> > > > +
> > > > +    input.guest_uaddr = (uint64_t)hva;
> > > 
> > > Thanks for changing these; although it fails a 32bit build (which
> > > is probably mostly pointless for SEV, but it fails the build
> > > rather than building it out). The easy fix here seems to be:
> > >    (uint64_t)(uintptr_t)
> > 
> > That's a pointer width issue.  The recommended way to communicate
> > to the compiler that we really want to cast a 32 bit pointer to a
> > 64 bit value is actually to cast to unsigned long before casting to
> > pointer, so
> > 
> > (uint64_t)(unsigned long)hva
> > 
> > Many other things work, of course, but if you follow the
> > recommendation you (hopefully) don't trip future compiler warnings.
> 
> OK, fair enough
> (Out of curiosity can you explain why unsigned long not uintptr_t?)

You have to cast it to a matching variable size integer first, so both
unsigned long and uintptr_t work because they're 64 bits on LP64 and 32
bits on ILP32.  The only minor problem with uintptr_t is that the
checker may or may not be programmed to know what it does and there's
nothing more annoying than trying to get a warning message you know to
be wrong fixed.

James