From mboxrd@z Thu Jan  1 00:00:00 1970
From: Eric Lacombe <goretux@gmail.com>
Subject: Re: [x86] - technical questions about HV implementation on Intel VT
Date: Tue, 21 Apr 2009 13:16:06 +0200
Message-ID: <200904211316.06778.goretux@gmail.com>
References: <200903241822.11529.goretux@gmail.com> <200904201953.53815.goretux@gmail.com> <49ECBD5D.4080205@redhat.com>
Mime-Version: 1.0
Content-Type: Text/Plain; charset=iso-8859-15
Content-Transfer-Encoding: QUOTED-PRINTABLE
Cc: kvm@vger.kernel.org
To: Avi Kivity <avi@redhat.com>
Return-path: <kvm-owner@vger.kernel.org>
Received: from mail-bw0-f163.google.com ([209.85.218.163]:49954 "EHLO
	mail-bw0-f163.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1751352AbZDULQM convert rfc822-to-8bit (ORCPT
	<rfc822;kvm@vger.kernel.org>); Tue, 21 Apr 2009 07:16:12 -0400
Received: by bwz7 with SMTP id 7so1698070bwz.37
        for <kvm@vger.kernel.org>; Tue, 21 Apr 2009 04:16:09 -0700 (PDT)
In-Reply-To: <49ECBD5D.4080205@redhat.com>
Content-Disposition: inline
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

Thanks a lot for your answer ;)

Le lundi 20 avril 2009 20:22:21 Avi Kivity, vous avez =E9crit :
> Eric Lacombe wrote:
[...]
> echo and pwd are part of bash, so they are probably in memory.  I gue=
ss
> once you go to disk things fail.
>
> Try to boot the entire OS from initramfs (and keep it there).

I will try this but maybe what follows say that the problem is elsewher=
e.

[...]
> > (Recall: When loaded, my module use VT-x to go on vmx root operatio=
n,
> > then it creates a vmcs in order to execute the OS inside a VM.)
>
> I imagine you have interrupts working properly?  Does 'watch -d cat
> /proc/interrupts' give the expected results (run it before you enter =
vmx
> to load it into cache)?

I made many times this test:
1. I run on a first console 'watch -n0,2 -d cat /proc/interrupts'
2. I load from another console my module (that is modified at the begin=
ning of=20
its init step with the addition of a schedule_timeout_uninterruptible()=
)
3. I switch to the first console and wait for schedule_timeout to retur=
n

And when my module does its stuff, the machine freezes... Maybe my modu=
le =20
"implies" a deadlock in the VFS after a VM-entry?

Note: in my module, I do not intercept (in the VM-execution controls) e=
xternal=20
interruptions nor exception.

I also check in the documentation the "VMX aborts" but it does not seem=
 to be=20
my problem --- the freeze occurs when I do not use MSR load/store areas=
 as=20
well as when I use them.
Note: now, I just load/store the Kernel_GS_BASE MSR to cover swapgs, ev=
en if=20
it is not actually necessary for my module. (Besides, I intercept rdmsr=
/wrmsr=20
as can be seen in the logs, and modify accordingly the VMCS when needed=
).

>
> Are you virtualizing memory, or does the guest manipulate page tables
> directly?

The guest manipulates directly page tables in my current module.
I just handle two cases of cr3 access ("mov from" and "mov to") by just=
=20
carrying out the "mov" in the guest registers.

Best regards

	Eric Lacombe