Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.33 as permitted sender) client-ip=23.128.96.33;
Date:   Fri, 3 Nov 2023 16:12:16 -0700
In-Reply-To: <8c6f06ae-d1d3-40ea-9bed-8ca949eaff5f@xen.org>
Mime-Version: 1.0
References: <20231002095740.1472907-1-paul@xen.org> <20231002095740.1472907-6-paul@xen.org>
 <ZUGScpSFlojjloQk@google.com> <8c6f06ae-d1d3-40ea-9bed-8ca949eaff5f@xen.org>
Message-ID: <ZUV-UG-Tm6HREWi2@google.com>
Subject: Re: [PATCH v7 05/11] KVM: pfncache: allow a cache to be activated
 with a fixed (userspace) HVA
From:   Sean Christopherson <seanjc@google.com>
To:     paul@xen.org
Cc:     kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
        Paul Durrant <pdurrant@amazon.com>,
        David Woodhouse <dwmw@amazon.co.uk>,
        Paolo Bonzini <pbonzini@redhat.com>,
        David Woodhouse <dwmw2@infradead.org>
Content-Type: text/plain; charset="us-ascii"
Precedence: bulk

On Thu, Nov 02, 2023, Paul Durrant wrote:
> On 31/10/2023 23:49, Sean Christopherson wrote:
> > On Mon, Oct 02, 2023, Paul Durrant wrote:
> > > diff --git a/include/linux/kvm_types.h b/include/linux/kvm_types.h
> > > index 6f4737d5046a..d49946ee7ae3 100644
> > > --- a/include/linux/kvm_types.h
> > > +++ b/include/linux/kvm_types.h
> > > @@ -64,7 +64,7 @@ struct gfn_to_hva_cache {
> > >   struct gfn_to_pfn_cache {
> > >   	u64 generation;
> > > -	gpa_t gpa;
> > > +	u64 addr;
> > 
> > Holy moly, we have unions for exactly this reason.
> > 
> > 	union {
> > 		gpa_t gpa;
> > 		unsigned long addr;
> > 	};
> > 
> > But that's also weird and silly because it's basically the exact same thing as
> > "uhva".  If "uhva" stores the full address instead of the page-aligned address,
> > then I don't see a need for unionizing the gpa and uhva.
> 
> Ok, I think that'll be more invasive but I'll see how it looks.

Invasive is fine.  Not ideal, but fine.  If the resulting code is a mess, then
that's a problem, but churn in and of itself isn't awful if the end result is a
net positive.

> > kvm_xen_vcpu_get_attr() should darn well explicitly check that the gpc stores
> > the correct type and not bleed ABI into the gfn_to_pfn_cache implementation.
> 
> I guess if we leave gpa alone and make it INVALID_GPA for caches initialized
> using an HVA then that can be checked. Is that what you mean here?

Yep, that should work.