Re: [Qemu-devel] [PULL 02/15] migration: extend VMStateInfo

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

* Cornelia Huck (cornelia.huck@de.ibm.com) wrote:
> On Wed, 25 Jan 2017 12:00:53 +0000
> "Dr. David Alan Gilbert" <dgilbert@redhat.com> wrote:
> 
> > * Fam Zheng (famz@redhat.com) wrote:
> > > On Tue, 01/24 18:47, Dr. David Alan Gilbert (git) wrote:
> > > > diff --git a/hw/intc/s390_flic_kvm.c b/hw/intc/s390_flic_kvm.c
> > > > index c313166..da8e4df 100644
> > > > --- a/hw/intc/s390_flic_kvm.c
> > > > +++ b/hw/intc/s390_flic_kvm.c
> > > > @@ -286,7 +286,8 @@ static void kvm_s390_release_adapter_routes(S390FLICState *fs,
> > > >   * increase until buffer is sufficient or maxium size is
> > > >   * reached
> > > >   */
> > > > -static void kvm_flic_save(QEMUFile *f, void *opaque, size_t size)
> > > > +static int kvm_flic_save(QEMUFile *f, void *opaque, size_t size,
> > > > +                         VMStateField *field, QJSON *vmdesc)
> > > >  {
> > > >      KVMS390FLICState *flic = opaque;
> > > >      int len = FLIC_SAVE_INITIAL_SIZE;
> > > > @@ -319,6 +320,8 @@ static void kvm_flic_save(QEMUFile *f, void *opaque, size_t size)
> > > >                          count * sizeof(struct kvm_s390_irq));
> > > >      }
> > > >      g_free(buf);
> > > > +
> > > > +    return 0;
> > > >  }
> > > 
> > > This hunk left one 'return' behind in the function, which should have been
> > > changed to 'return 0' as well, and now the compiler is not happy:
> > > 
> > > /var/tmp/patchew-tester-tmp-itftfkl9/src/hw/intc/s390_flic_kvm.c: In function ‘kvm_flic_save’:
> > > /var/tmp/patchew-tester-tmp-itftfkl9/src/hw/intc/s390_flic_kvm.c:306:9: error: ‘return’ with no value, in function returning non-void [-Werror]
> > >          return;
> > >          ^~~~~~
> > > /var/tmp/patchew-tester-tmp-itftfkl9/src/hw/intc/s390_flic_kvm.c:289:12: note: declared here
> > >  static int kvm_flic_save(QEMUFile *f, void *opaque, size_t size,
> > >             ^~~~~~~~~~~~~
> > > cc1: all warnings being treated as errors
> > 
> > OK, so it looks like that's a failure path, adding a return -ENOMEM would seem to make
> > sense there.
> 
> Just saw this. I don't think we want -ENOMEM, as that would change the
> actual state being saved, no?

But isn't that the intention of this function?

    buf = g_try_malloc0(len);
    if (!buf) {
        /* Storing FLIC_FAILED into the count field here will cause the
         * target system to fail when attempting to load irqs from the
         * migration state */
        error_report("flic: couldn't allocate memory");
        qemu_put_be64(f, FLIC_FAILED);
        return;
    }

What should happen on the destination - should the migration fail?
If we want the migration to fail then we should now return an error
status rather than 0, and then we see a failed migration on the source
as well.

Dave

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