Re: [Qemu-devel] [PATCH v2] hmp: gpa2hva and gpa2hpa hostaddr command

["Dr. David Alan Gilbert" <dgilbert@redhat.com>]

* Paolo Bonzini (bonzini@gnu.org) wrote:
> 
> 
> On 20/03/2017 18:16, Paolo Bonzini wrote:
> > 
> > 
> > On 20/03/2017 18:01, Markus Armbruster wrote:
> >> Peter Maydell <peter.maydell@linaro.org> writes:
> >>
> >>> On 20 March 2017 at 16:29, Markus Armbruster <armbru@redhat.com> wrote:
> >>>> Peter Maydell <peter.maydell@linaro.org> writes:
> >>>>> I have some comments which feel kind of nit-picky, but since this
> >>>>> is a public-facing HMP API I think they need attention since we only
> >>>>> get one chance to get it right.
> >>>>
> >>>> HMP is not a stable interface.  If we get it wrong, we change it.  If
> >>>> our change breaks your usage, you get to keep the pieces.
> >>>
> >>> Oh yes, I had my HMP and QMP the wrong way round.
> >>>
> >>> ...which reminds me that I thought we preferred all HMP commands
> >>> to be implemented in terms of their QMP equivalent ?
> >>
> >> Yes.  We make exceptions for commands we believe have no QMP use, such
> >> as "print".  I figure the rationale for these is "just for testing".
> >> Paolo, can you confirm?
> > 
> > Yes.  If somebody comes up with a non-testing use, I suppose we can
> > always add a QMP variant.  I'll send v3 which uses the current monitor
> > CPU's address space according to Peter's review.
> 
> Since there is no CPU method to transform a CPU state into a MemTxAttrs
> value, and the MemTxAttrs are needed to get the right address space, v2
> is the best I can do for 2.9.
> 
> Changing it to take the CPU state into account (e.g. include secure
> memory when in secure mode) can be done in later releases if necessary.
> 
> David, are you queuing the patch?
> 
> Paolo

Unfortunately it's hitting a warning on the mingw build:

/home/dgilbert/git/hmp/monitor.c: In function 'hmp_gpa2hva':
/home/dgilbert/git/hmp/monitor.c:1461:5: error: 'mr' may be used uninitialized in this function [-Werror=maybe-uninitialized]
     memory_region_unref(mr);
     ^~~~~~~~~~~~~~~~~~~~~~~

I think that's the compiler not being clever enough to spot that path
would always be initialised.

I guess it just needs an = NULL on the mr.

(i686-w64-mingw32-gcc gcc version 6.3.0 20161221 (Fedora MinGW 6.3.0-1.fc25) (GCC) )

Dave
--
Dr. David Alan Gilbert / dgilbert@redhat.com / Manchester, UK