Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 27 Oct 2022 15:10:07 +0000
From:   Sean Christopherson <seanjc@google.com>
To:     Paolo Bonzini <pbonzini@redhat.com>
Cc:     kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
        Michal Luczaj <mhal@rbox.co>,
        David Woodhouse <dwmw2@infradead.org>
Subject: Re: [PATCH v2 03/16] KVM: x86: Always use non-compat
 vcpu_runstate_info size for gfn=>pfn cache
Message-ID: <Y1qfT2Mnwk5GmjFI@google.com>
References: <20221013211234.1318131-1-seanjc@google.com>
 <20221013211234.1318131-4-seanjc@google.com>
 <afad5f40-03ef-1380-9bfe-03bbaaed47a9@redhat.com>
 <Y1qZagwM0dMBjYhe@google.com>
 <24768aa3-0e2e-6d29-2749-9d74a26f9205@redhat.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <24768aa3-0e2e-6d29-2749-9d74a26f9205@redhat.com>
Precedence: bulk

On Thu, Oct 27, 2022, Paolo Bonzini wrote:
> On 10/27/22 16:44, Sean Christopherson wrote:
> > > - long mode cannot be changed after the shared info page is enabled (which
> > > makes sense because the shared info page also has a compat version)
> > 
> > How is this not introducing an additional restriction?  This seems way more
> > onerous than what is effectively a revert.
> > 
> > > - the caches must be activated after the shared info page (which enforces
> > > that the vCPU attributes are set after the VM attributes)
> > > 
> > > This is technically a userspace API break, but nobody is really using this
> > > API outside Amazon so...  Patches coming after I finish testing.
> > 
> > It's not just userspace break, it affects the guest ABI as well.
> 
> Yes, I was talking of the VMM here; additional restrictions are fine there.

Additional restrictions are fine where?

> The guests however should be compatible with Xen, so you also need to
> re-activate the cache after the hypercall page is written, but that's two
> lines of code.

And do what if the guest transitions from 32-bit => 64-bit and the cache isn't
aligned for 64-bit?  E.g. kvm_xen_set_evtchn() will silently drop events no matter
what KVM does.  In other words, I don't see how KVM can provide a same ABI without
forcing the cached pages to be aligned for the largets possible size.