Re: linux-next: Tree for Jan 24 (kvm)

[Sean Christopherson <sean.j.christopherson@intel.com>]

On Fri, Jan 24, 2020 at 01:48:07PM -0800, Jim Mattson wrote:
> On Fri, Jan 24, 2020 at 1:30 PM Sean Christopherson
> <sean.j.christopherson@intel.com> wrote:
> >
> > On Fri, Jan 24, 2020 at 12:51:31PM -0800, Randy Dunlap wrote:
> > > On 1/23/20 10:33 PM, Stephen Rothwell wrote:
> > > > Hi all,
> > > >
> > > > Changes since 20200123:
> > > >
> > > > The kvm tree gained a conflict against Linus' tree.
> > > >
> > >
> > > on i386:
> > >
> > > ../arch/x86/kvm/x86.h:363:16: warning: right shift count >= width of type [-Wshift-count-overflow]
> >
> > Jim,
> >
> > This is due to using "unsigned long data" for kvm_dr7_valid() along with
> > "return !(data >> 32);" to check for bits being set in 63:32.  Any
> > objection to fixing the issue by making @data a u64?  Part of me thinks
> > that's the proper behavior anyways, i.e. the helper is purely a reflection
> > of the architectural requirements, the caller is responsible for dropping
> > bits appropriately based on the current mode.
> 
> Why not just change that bad return statement to one of the
> alternatives you had suggested previously?

Because it's not consistent with e.g. is_noncanonical_address() and I don't
like dropping bits 63:32 of vmcs12->guest_dr7 when kvm_dr7_valid() is called
from nested_vmx_check_guest_state().  KVM will eventually drop the bits
anyways when propagating vmcs12->guest_dr7 to vmcs02, but I'd prefer the
consistency check to not rely on that behavior.

> I think "return !(data >> 32)" was the only suggested alternative that
> doesn't work. :-)