Subject: Re: [PART1 RFC 5/9] svm: Add VMEXIT handlers for AVIC
To: Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>,
        =?UTF-8?B?UmFkaW0gS3LEjW3DocWZ?= <rkrcmar@redhat.com>
References: <56C2C1BF.7010700@amd.com> <56C312E1.1080902@redhat.com>
 <20160216141330.GG10555@potion.brq.redhat.com> <56C354A5.4040807@redhat.com>
 <20160216180618.GA18952@potion.brq.redhat.com> <56C52B80.5050104@amd.com>
 <20160218141817.GA6289@potion.brq.redhat.com> <56C5DA62.8080204@redhat.com>
 <20160218154343.GA18904@potion.brq.redhat.com> <56C5E8E1.9060900@redhat.com>
 <20160218162701.GB6289@potion.brq.redhat.com> <56C5FCCF.5010802@redhat.com>
 <56D8150E.1070408@amd.com>
Cc: joro@8bytes.org, alex.williamson@redhat.com, gleb@kernel.org,
        kvm@vger.kernel.org, linux-kernel@vger.kernel.org, wei@redhat.com,
        sherry.hurwitz@amd.com, "Wu, Feng" <feng.wu@intel.com>
From: Paolo Bonzini <pbonzini@redhat.com>
Message-ID: <56D816F0.2040403@redhat.com>
Date: Thu, 3 Mar 2016 11:50:24 +0100
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101
 Thunderbird/38.6.0
MIME-Version: 1.0
In-Reply-To: <56D8150E.1070408@amd.com>
Content-Type: text/plain; charset=utf-8
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 883
Lines: 24


On 03/03/2016 11:42, Suravee Suthikulpanit wrote:
> In facts, instead of setting up the vAPIC backing page address when
> calling kvm_arch_vcpu_load(), we should be able to do it when calling
> kvm_arch_vcpu_sched_in(). This seems more appropriate since the
> kvm_arch_vcpu_load() is also called in many unnecessary occasions via
> vcpu_load() (in the arch/x86/kvm/x86.c). The same goes for the
> kvm_arch_vcpu_put().
> 
> However, there is no kvm_arch_vcpu_sched_out(). But that can be added
> easily.
> 
> What do you think?

How much code is involved?  It seems an unnecessary complication...
Unless you can profile a difference, I would leave it in
kvm_arch_vcpu_load()/kvm_arch_vcpu_put().

Note that sched_in and sched_out are not called once per invocation of
KVM_RUN---only through the preempt notifiers.  So I'm not sure it is
enough to use sched_in and sched_out.

Paolo