Received-SPF: pass (google.com: domain of linux-kernel+bounces-171780-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.48.161 as permitted sender) client-ip=147.75.48.161;
Precedence: bulk
MIME-Version: 1.0
References: <20240404185034.3184582-1-pbonzini@redhat.com> <20240404185034.3184582-7-pbonzini@redhat.com>
 <ZiZBjtQvUuuqqKNF@yilunxu-OptiPlex-7050>
In-Reply-To: <ZiZBjtQvUuuqqKNF@yilunxu-OptiPlex-7050>
From: Paolo Bonzini <pbonzini@redhat.com>
Date: Tue, 7 May 2024 18:17:49 +0200
Message-ID: <CABgObfaULPdNbf8yZHEwesDx+KWvt1A+Eps4xY4DkgzhTq9AzA@mail.gmail.com>
Subject: Re: [PATCH 06/11] KVM: guest_memfd: Add hook for initializing memory
To: Xu Yilun <yilun.xu@linux.intel.com>
Cc: linux-kernel@vger.kernel.org, kvm@vger.kernel.org, seanjc@google.com, 
	michael.roth@amd.com, isaku.yamahata@intel.com
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable

On Mon, Apr 22, 2024 at 12:58=E2=80=AFPM Xu Yilun <yilun.xu@linux.intel.com=
> wrote:
> > In some cases, it is necessary to defer the preparation of the pages to
> > handle things like in-place encryption of initial guest memory payloads
> > before marking these pages as 'private'/'guest-owned'.  Add an argument
> > (always true for now) to kvm_gmem_get_folio() that allows for the
> > preparation callback to be bypassed.  To detect possible issues in
>
> IIUC, we have 2 dedicated flows.
> 1 kvm_gmem_get_pfn() or kvm_gmem_allocate()
>   a. kvm_gmem_get_folio()
>   b. gmem_prepare() for RMP
>
> 2 in-place encryption or whatever
>   a. kvm_gmem_get_folio(FGP_CREAT_ONLY)
>   b. in-place encryption
>   c. gmem_prepare() for RMP
>
> Could we move gmem_prepare() out of kvm_gmem_get_folio(), then we could
> have straightforward flow for each case, and don't have to have an
> argument to pospone gmem_prepare().

There are 3 flows as you note above - kvm_gmem_get_pfn() and
kvm_gmem_allocate() are different paths but they all need to call the
prepare hook. It is a tempting idea to pull kvm_gmem_prepare_folio()
to the two functions (get_pfn and allocate) but the resulting code is
really ugly due to folio_unlock/folio_put.

> > -static struct folio *kvm_gmem_get_folio(struct inode *inode, pgoff_t i=
ndex)
> > +#ifdef CONFIG_HAVE_KVM_GMEM_PREPARE
> > +bool __weak kvm_arch_gmem_prepare_needed(struct kvm *kvm)
> > +{
> > +     return false;
> > +}
> > +#endif
>
> In which case HAVE_KVM_GMEM_PREPARE is selected but
> gmem_prepare_needed() is never implemented?  Then all gmem_prepare stuff
> are actually dead code.  Maybe we don't need this weak stub?

It's not needed indeed.

> > +     if (prepare) {
> > +             int r =3D kvm_gmem_prepare_folio(inode, index, folio);
> > +             if (r < 0) {
> > +                     folio_unlock(folio);
> > +                     folio_put(folio);
> > +                     return ERR_PTR(r);
> > +             }
> > +     }
> > +
>
> Do we still need to prepare the page if it is hwpoisoned? I see the
> hwpoisoned check is outside, in kvm_gmem_get_pfn().

Yep, it can be moved here.

Paolo