On Thu 11/29/07 at 15:17 PM, Nils.Nieuwejaar at Sun.COM wrote:
> On Thu 11/29/07 at 12:01 PM, Tim.Marsland at Sun.COM wrote:
> > >If we determine we are running in an hvm domain on a supported  version
> > >of Xen, we will add /platform/i86hvm/kernel to the beginning of the
> > >default search path.  Otherwise, we will leave the search path
> > >untouched, and Solaris will access all devices through the (possibly
> > >emulated)  physical interfaces.
> > 
> > But up until now (my understanding was) we didn't have to touch the
> > kernel to make these PV drivers work.  There's considerable advantage in
> > the  perception that we're just adding drivers, rather than patching the
> > kernel ..
> > 
> > (Were you *already* patching the kernel and I just didn't realize it?)
> 
> We can deliver the PV drivers without making any kernel modifications.
> Providing tranparent access to those drivers through the existing physical
> interfaces does require some minor kernel modifications.  Since we're
> already going to depend on a kernel patch to deliver these in an update, we
> decided to use this cleaner i86hvm approach, rather than hanging a PV bag
> off the side of the existing cmdk driver. 

Upon further reflection, by including new versions of cmdk and cmlb that
install in /platform/i86pc rather than /kernel, we should be able to
deliver all of this as a simple pkgadd for an update release. 

These modifications to cmdk and cmlb would not be something we want
integrated into ON.  They would just be a stopgap measure to simplify the
end-user experience for update releases that are already in the field.
Upgrading domU to the next update release would remove these modified
drivers, and would install a kernel with the necessary i86hvm support built
in.

Nils


Reply via email to