[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20200229183830.GB22451@linux.intel.com>
Date: Sat, 29 Feb 2020 10:38:30 -0800
From: Sean Christopherson <sean.j.christopherson@...el.com>
To: Paolo Bonzini <pbonzini@...hat.com>
Cc: Vitaly Kuznetsov <vkuznets@...hat.com>,
Wanpeng Li <wanpengli@...cent.com>,
Jim Mattson <jmattson@...gle.com>,
Joerg Roedel <joro@...tes.org>, kvm@...r.kernel.org,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH 48/61] KVM: x86: Do host CPUID at load time to mask KVM
cpu caps
On Tue, Feb 25, 2020 at 01:08:43PM -0800, Sean Christopherson wrote:
> On Tue, Feb 25, 2020 at 04:18:12PM +0100, Paolo Bonzini wrote:
> > On 01/02/20 19:52, Sean Christopherson wrote:
> > > +#ifdef CONFIG_KVM_CPUID_AUDIT
> > > + /* Entry needs to be fully populated when auditing is enabled. */
> > > + entry.function = cpuid.function;
> > > + entry.index = cpuid.index;
> > > +#endif
> >
> > This shows that the audit case is prone to bitrot, which is good reason
> > to enable it by default.
>
> I have no argument against that, especially since I missed this case during
> development and only caught it when running on a different system that I
> had happened to configure with CONFIG_KVM_CPUID_AUDIT=y. :-)
I ended up dropping the audit code altogether. The uops overhead wasn't
bad, but the code bloat was pretty rough, ~16 bytes per instance. The
final nail in the coffin was that the auditing would trigger false
positives if userspace configured CPUID leafs with a non-signficant index
to have a non-zero index, e.g. is_matching_cpuid_entry() ignores the index
if KVM_CPUID_FLAG_SIGNIFCANT_INDEX isn't set.
Powered by blists - more mailing lists