On Fri, Sep 15, 2017 at 06:40:40PM +0200, Paolo Bonzini wrote:
> On 15/09/2017 01:31, Jim Mattson wrote:
> > When the APIC virtualization address references unbacked memory, kvm
> > incorrectly fails a VM-entry with "invalid control field(s)." Until
> > this can be fixed, just skip the VMX control field tests that populate
> > a VMCS field with a physical address that isn't backed.
> >
> > Tested: <multi line explanation of how the code was tested>
> >
> > Effort: <Your effort group>
> > Google-Bug-Id: <buganizer-id>
> > Change-Id: I624091cb5cd7cfaae887de8f017e18b5f95d8f61
Nice Google commit message template :-)
> > Signed-off-by: Jim Mattson <jmattson@xxxxxxxxxx>
> > ---
> > x86/vmx_tests.c | 6 ++++++
> > 1 file changed, 6 insertions(+)
> >
> > diff --git a/x86/vmx_tests.c b/x86/vmx_tests.c
> > index 088d32aaee0b..e82bc04d72e2 100644
> > --- a/x86/vmx_tests.c
> > +++ b/x86/vmx_tests.c
> > @@ -3359,6 +3359,12 @@ static void test_vmcs_page_addr(const char *name,
> > bool ignored,
> > u64 addr)
> > {
> > + /*
> > + * Skip tests that reference unbacked memory for now, because
> > + * kvm doesn't always handle this situation correctly.
>
> Since in practice it's only 16 tests that fail, it would be nice to only
> XFAIL those. The following is a bit intrusive, but it works. What do
> you think?
>
A nice thing about using xfail=true|false now, is that it may later be
assigned to !ERRATA(...) in order to start expecting the tests to pass
when run on hosts that have the fix.
drew
