Re: [Qemu-devel] [PATCH 2/3] geometry detection: use HDIO_GETGEO

[Alexander Graf]

On 05/02/2012 12:27 PM, Christian Borntraeger wrote:
> On 27/04/12 18:12, Paolo Bonzini wrote:
> > Il 26/04/2012 15:49, Christian Borntraeger ha scritto:
> > > +#ifdef __linux__
> > > +    } else if (bdrv_ioctl(bs, HDIO_GETGEO,&geo) == 0) {
> > > +        *pcyls = geo.cylinders;
> > > +        *pheads = geo.heads;
> > > +        *psecs = geo.sectors;
> > > +        bdrv_set_geometry_hint(bs, *pcyls, *pheads, *psecs);
> > > +#endif
> > Perhaps you could instead move guess_disk_lchs to target-specific code,
> > adding add this code to the s390-specific implementation and under
> > #ifdef __s390__.  For x86 it doesn't make much sense, because a disk's
> > geometry most likely will be a wrong guess for the geometry that a guest
> > (for guests that care at all about geometries).
> Fine with me. We care about the geometry only for dasd devices, Even for 
> FCP-based
> SCSI devices on s390 the geometry is not relevant. So moving that part to
> s390 specific code might make sense if nobody else needs that.
> Is that the case?
> Alex, would that be ok for you?

As hinted in my other mail, I think the way to go would be to give a 
hint to the geometry code that we're running on a DASD disk. Then we can

  * Ask the host device if it can give us its geometry, if so, use it
  * Guess depending on the logical block size

and everyone should be happy :). I would really like to have as little 
#ifdef TARGET_S390 code in QEMU. And #ifdef __s390__ is even worse, as 
it means we won't be able to execise that code path on other architectures.


Alex