From mboxrd@z Thu Jan  1 00:00:00 1970
From: Radim =?utf-8?B?S3LEjW3DocWZ?= <rkrcmar@redhat.com>
Subject: Re: [kvm-unit-tests PATCH 2/3] x86/vmx: fix detection of unmapped PTE
Date: Mon, 3 Jul 2017 18:42:19 +0200
Message-ID: <20170703164219.GA30880@potion>
References: <20170629172647.22188-1-rkrcmar@redhat.com>
 <20170629172647.22188-3-rkrcmar@redhat.com>
 <ce8586df-d26b-6b2e-5335-d523dea2370e@redhat.com>
 <83f91b73-f0ae-3e78-fc90-a79fd31fd574@redhat.com>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Cc: kvm@vger.kernel.org, David Matlack <dmatlack@google.com>,
        Peter Feiner <pfeiner@google.com>
To: Paolo Bonzini <pbonzini@redhat.com>
Return-path: <kvm-owner@vger.kernel.org>
Received: from mx1.redhat.com ([209.132.183.28]:42878 "EHLO mx1.redhat.com"
        rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
        id S932087AbdGCQmX (ORCPT <rfc822;kvm@vger.kernel.org>);
        Mon, 3 Jul 2017 12:42:23 -0400
Content-Disposition: inline
In-Reply-To: <83f91b73-f0ae-3e78-fc90-a79fd31fd574@redhat.com>
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

2017-07-03 12:34+0200, Paolo Bonzini:
> On 30/06/2017 12:33, Paolo Bonzini wrote:
> > This breaks the vmx PML test:
> > 
> > Test suite: PML
> > PASS: PML - Dirty GPA Logging
> > ERROR - unexpected stage, 2.
> > VMEXIT info:
> > 	vmexit reason = 18
> > 	exit qualification = 0
> > 	Bit 31 of reason = 0
> > 	guest_rip = 0x4049b3
> > 	RAX=0x149a000    RBX=0x3f8    RCX=0x184    RDX=0x3fd
> > 	RSP=0x46efcf    RBP=0x46efdf    RSI=0x41a8a3    RDI=0x41a8a3
> > 	R8 =0xa    R9 =0x3f8    R10=0    R11=0
> > 	R12=0    R13=0    R14=0    R15=0
> > 
> > 
> > (should have been "PASS: PML Full Event").  I didn't investigate why.
> 
> Hmm, actually it seems to come and go.  I've applied patches 2 and 3 for
> now.

I had to use a different machine to reproduce ... with the patches, the
guest report() (probably) accesses 1 more page, which adds one entry to
PML and changes the progress of this pml_main loop:

	while (vmx_get_test_stage() == 1) {
		*((u32 *)data_page2) = 0x1;
		if (count++ > PML_INDEX)
			break;
		vmcall();
	}

Each cycle adds 4 entries to the PML and the extra entry makes the log
fill just before the vmcall.  I don't think the one extra access is
something we need to tend to, so I'll send a fix for the PML test.