From mboxrd@z Thu Jan 1 00:00:00 1970 From: "Michael S. Tsirkin" Subject: Re: [PATCH for qemu-kvm-1.2] pci-assign: Use proper pin number for INTx-to-IRQ routing Date: Wed, 29 Aug 2012 16:02:07 +0300 Message-ID: <20120829130207.GA6506@redhat.com> References: <503DFCB8.8000407@siemens.com> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Cc: Avi Kivity , Marcelo Tosatti , kvm , Alex Williamson To: Jan Kiszka Return-path: Received: from mx1.redhat.com ([209.132.183.28]:50168 "EHLO mx1.redhat.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752663Ab2H2NAv (ORCPT ); Wed, 29 Aug 2012 09:00:51 -0400 Content-Disposition: inline In-Reply-To: <503DFCB8.8000407@siemens.com> Sender: kvm-owner@vger.kernel.org List-ID: On Wed, Aug 29, 2012 at 01:27:52PM +0200, Jan Kiszka wrote: > Broken by commit e21f28b497. > > Signed-off-by: Jan Kiszka > --- > hw/kvm/pci-assign.c | 2 +- > 1 files changed, 1 insertions(+), 1 deletions(-) > > diff --git a/hw/kvm/pci-assign.c b/hw/kvm/pci-assign.c > index 9cce02c..fea05b4 100644 > --- a/hw/kvm/pci-assign.c > +++ b/hw/kvm/pci-assign.c > @@ -892,7 +892,7 @@ static int assign_intx(AssignedDevice *dev) > pci_device_set_intx_routing_notifier(&dev->dev, > assigned_dev_update_irq_routing); > > - intx_route = pci_device_route_intx_to_irq(&dev->dev, 0); > + intx_route = pci_device_route_intx_to_irq(&dev->dev, dev->intpin); > assert(intx_route.mode != PCI_INTX_INVERTED); > > if (dev->intx_route.mode == intx_route.mode && Hmm but looking at it, I see a different problem: /* handle interrupt routing */ e_intx = dev->dev.config[0x3d] - 1; dev->intpin = e_intx; PCI spec however says: PCI defines one interrupt line for a single function device and up to four interrupt lines for a multi-function6 device or connector. For a single function device, only INTA# may be used while the other three interrupt lines have no meaning. And: When several independent functions are integrated into a single device, it will be referred to as a multi- function device. Each function on a multi-function device has its own configuration space. So this is not easily virtualizeable. I think in practice it should be only legal to assign VFs or all functions of a physical device in one go. In last case, we can then make sure function numbers match. Alex? > -- > 1.7.3.4