From mboxrd@z Thu Jan  1 00:00:00 1970
From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Subject: Re: [PATCH 3/3] drivers/vfio/pci: Fix MSIx message lost
Date: Mon, 03 Mar 2014 16:00:49 +1100
Message-ID: <1393822849.10727.26.camel@pasglop>
References: <1393817042-13758-1-git-send-email-shangw@linux.vnet.ibm.com>
	 <1393817042-13758-3-git-send-email-shangw@linux.vnet.ibm.com>
	 <1393818710.10727.22.camel@pasglop> <5314086F.10108@ozlabs.ru>
Mime-Version: 1.0
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 7bit
Cc: Gavin Shan <shangw@linux.vnet.ibm.com>, kvm@vger.kernel.org,
	alex.williamson@redhat.com
To: Alexey Kardashevskiy <aik@ozlabs.ru>
Return-path: <kvm-owner@vger.kernel.org>
Received: from gate.crashing.org ([63.228.1.57]:39255 "EHLO gate.crashing.org"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1751444AbaCCFBP (ORCPT <rfc822;kvm@vger.kernel.org>);
	Mon, 3 Mar 2014 00:01:15 -0500
In-Reply-To: <5314086F.10108@ozlabs.ru>
Sender: kvm-owner@vger.kernel.org
List-ID: <kvm.vger.kernel.org>

On Mon, 2014-03-03 at 15:43 +1100, Alexey Kardashevskiy wrote:
> While it works for our particular problem and seems correct, it has one
> flaw - hw/pci/msix.c will not generate this backtrace if masking bit does
> not change which can happen in general:
> ===
> static void msix_handle_mask_update(PCIDevice *dev, int vector, bool
> was_masked)
> {
>     bool is_masked = msix_is_masked(dev, vector);
> 
>     if (is_masked == was_masked) {
>         return;
>     }
> ===
> 
> Or if masking bit is the same, nothing bad is expected?...

Hrm ok, so it will work in this specific case but might not in the general
case of a driver triggering some kind of local reset on the device requiring
the MSI-X to be restored. The guest will write but qemu will swallow them ...

I think that needs to be fixed but it might be hard without introducing
a new ioctl from what I can see of the way the code is structured... Unless
qemu turns that into a disable/enable pair I suppose.

Cheers,
Ben.