Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Message-ID: <1510757432.21121.294.camel@redhat.com>
Subject: Re: [PATCH] x86,kvm: move qemu/guest FPU switching out to vcpu_run
From:   Rik van Riel <riel@redhat.com>
To:     Paolo Bonzini <pbonzini@redhat.com>,
        David Hildenbrand <david@redhat.com>
Cc:     kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
        tglx@linutronix.de, rkrcmar@redhat.com, borntraeger@de.ibm.com
Date:   Wed, 15 Nov 2017 09:50:32 -0500
In-Reply-To: <d1aa3e79-7538-6671-dbff-04a3b000bfe5@redhat.com>
References: <20171114001223.441ea2ca@annuminas.surriel.com>
         <a3c862c2-e5f9-f394-885c-10fde7904f03@redhat.com>
         <1510682877.30057.1.camel@redhat.com>
         <6f73cff1-eddf-7dc6-a93c-31c20e8520d7@redhat.com>
         <d1aa3e79-7538-6671-dbff-04a3b000bfe5@redhat.com>
Organization: Red Hat, Inc
Content-Type: text/plain; charset="UTF-8"
Mime-Version: 1.0
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On Wed, 2017-11-15 at 09:34 +0100, Paolo Bonzini wrote:
> On 14/11/2017 20:40, David Hildenbrand wrote:
> > I think we should check all get/put_fpu callers if they need
> > preempt_disable().
> > 
> > E.g. em_fxrstor() needs disabled preemption as we temporarily
> > save + restore some host register (via fxsave + fxrstor) under some
> > circumstances that are not saved/restored when switching to/back
> > from
> > another process. We should double check.
> 
> Rik may correct me, but I believe that you don't need
> preempt_disable/enable because preempt notifiers do this for you.

We no longer even need the preempt notifiers to save and
restore the guest FPU state.

The context switch code itself will save the FPU state
from the registers, into current->thread.fpu.state, when
the VCPU thread gets scheduled out.

When the VCPU thread gets scheduled in, the scheduler
will restore the guest FPU state from current->thread.fpu.state.

At this point, vcpu->arch.guest_fpu may be OUT OF DATE.

However, this is just fine, because we will save the guest
FPU state into vcpu->arch.guest_fpu in kvm_put_guest_fpu,
before we leave the KVM_RUN ioctl, and before we release
the vcpu->mutex.

In other words, by the time anybody else can examine the
VCPU FPU state (after they obtain the vcpu->mutex), the
vcpu->arch.guest_fpu area will contain the correct FPU
state.

From 1584143680664312426@xxx Wed Nov 15 14:45:41 +0000 2017
X-GM-THRID: 1584017174910331026
X-Gmail-Labels: Inbox,Category Forums,HistoricalUnread