From mboxrd@z Thu Jan  1 00:00:00 1970
From: Marcelo Tosatti <mtosatti@redhat.com>
Subject: Re: [patch 2/5] KVM: MMU: allow pinning spte translations (TDP-only)
Date: Thu, 19 Jun 2014 16:22:57 -0300
Message-ID: <20140619192257.GA5561@amt.cnet>
References: <20140618231203.846608908@amt.cnet>
 <20140618231521.569025131@amt.cnet>
 <20140619072116.GC10948@minantech.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: kvm@vger.kernel.org, ak@linux.intel.com, pbonzini@redhat.com,
	xiaoguangrong@linux.vnet.ibm.com, avi@cloudius-systems.com
To: Gleb Natapov <gleb@kernel.org>
Return-path: <kvm-owner@vger.kernel.org>
Received: from mx1.redhat.com ([209.132.183.28]:44350 "EHLO mx1.redhat.com"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1757787AbaFSTXU (ORCPT <rfc822;kvm@vger.kernel.org>);
	Thu, 19 Jun 2014 15:23:20 -0400
Content-Disposition: inline
In-Reply-To: <20140619072116.GC10948@minantech.com>
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

On Thu, Jun 19, 2014 at 10:21:16AM +0300, Gleb Natapov wrote:
> On Wed, Jun 18, 2014 at 08:12:05PM -0300, mtosatti@redhat.com wrote:
> > Allow vcpus to pin spte translations by:
> > 
> > 1) Creating a per-vcpu list of pinned ranges.
> What if memory slot containing pinned range is going away?

->page_fault() should fail and guest abort. Will double check.

> > 2) On mmu reload request:
> > 	- Fault ranges.
> > 	- Mark sptes with a pinned bit.
> Should also be marked "dirty" as per SDM:
>  The three DS save area sections should be allocated from a non-paged pool, and marked accessed and dirty

This (SDM text) is about guest pagetable AFAICS.

> > +	mutex_unlock(&vcpu->arch.pinned_mmu_mutex);
> > +
> > +	if (vcpu->arch.nr_pinned_ranges >=
> > +	    KVM_MAX_PER_VCPU_PINNED_RANGE)
> > +		return -ENOSPC;
> Shouldn't we refuse to register pinned range if !TDP?

Sure.

> > +			/* MMU notifier sequence window: retry */
> > +			if (!r && !pinned)
> > +				kvm_make_request(KVM_REQ_MMU_RELOAD, vcpu);
> > +			if (r) {
> > +				kvm_make_request(KVM_REQ_TRIPLE_FAULT, vcpu);
> I do not think triple fault is appropriate here. The reasons for triple fault are
> documented in SDM and this is not one of them. What about error exit to user space?

Agree, will change.