Date:   Fri, 17 Mar 2023 00:18:41 +0000
In-Reply-To: <a026b6ddf62843129193842d80edd182@huawei.com>
Mime-Version: 1.0
References: <20230316154554.1237-1-shameerali.kolothum.thodi@huawei.com>
 <ZBNLnp7c1JvDsmHm@google.com> <6b9e8589281c4d2bae46eba36f77afe7@huawei.com>
 <ZBN0pFN/nF8G3fWl@google.com> <a026b6ddf62843129193842d80edd182@huawei.com>
Message-ID: <ZBOx4b1yUpnxu/I1@google.com>
Subject: Re: [PATCH] KVM: Add the missing stub function for kvm_dirty_ring_check_request()
From:   Sean Christopherson <seanjc@google.com>
To:     Shameerali Kolothum Thodi <shameerali.kolothum.thodi@huawei.com>
Cc:     "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        "kvm@vger.kernel.org" <kvm@vger.kernel.org>,
        "gshan@redhat.com" <gshan@redhat.com>,
        "maz@kernel.org" <maz@kernel.org>
Content-Type: text/plain; charset="us-ascii"
Precedence: bulk

On Thu, Mar 16, 2023, Shameerali Kolothum Thodi wrote:
> > From: Sean Christopherson [mailto:seanjc@google.com]
> > On Thu, Mar 16, 2023, Shameerali Kolothum Thodi wrote:
> > > > From: Sean Christopherson [mailto:seanjc@google.com] On Thu, Mar 16,
> > > > 2023, Shameer Kolothum wrote:
> > > > > The stub for !CONFIG_HAVE_KVM_DIRTY_RING case is missing.
> > > >
> > > > No stub is needed.  kvm_dirty_ring_check_request() isn't called from
> > > > common code, and should not (and isn't unless I'm missing something)
> > > > be called from arch code unless CONFIG_HAVE_KVM_DIRTY_RING=y.
> > > >
> > > > x86 and arm64 are the only users, and they both select
> > > > HAVE_KVM_DIRTY_RING unconditionally when KVM is enabled.
> > >
> > > Yes, it is at present not called from anywhere other than x86 and arm64.
> > > But I still think since it is a common helper, better to have a stub.
> > 
> > Why?  It buys us nothing other than dead code, and even worse it could let
> > a bug that would otherwise be caught during build time escape to run time.
> 
> Agree, it buys nothing now:) It just came up while I was playing with a custom
> build without HAVE_KVM_DIRTY_RING. Since all other functions there has a stub
> just thought it would make it easier for future common usage. We could very well
> leave it till that comes up as well.

Stubs are typically only added when they are strictly necessary.  Providing a stub
would make things "easier" in the sense that it could theoretically avoid a build
error, but as above, in many cases we _want_ build errors when new code behaves
in a way that diverges from what's expected/established.