On Thu, 2011-07-28 at 08:58 -0700, Rose, Gregory V wrote: > > -----Original Message----- > > From: Ian Campbell [mailto:ijc@xxxxxxxxxxxxxx] > > Sent: Thursday, July 28, 2011 8:11 AM > > To: Rose, Gregory V; Konrad Rzeszutek Wilk > > Cc: netdev@xxxxxxxxxxxxxxx; davem@xxxxxxxxxxxxx; > > bhutchings@xxxxxxxxxxxxxx; Kirsher, Jeffrey T; Jesse Barnes; linux- > > pci@xxxxxxxxxxxxxxx > > Subject: Re: [RFC net-next PATCH 1/4] pci: Add flag indicating device has > > been assigned by KVM > > > > On Wed, 2011-07-27 at 15:17 -0700, Greg Rose wrote: > > > Device drivers that create and destroy SR-IOV virtual functions via > > > calls to pci_enable_sriov() and pci_disable_sriov can cause catastrophic > > > failures if they attempt to destroy VFs while they are assigned to > > > guest virtual machines. By adding a flag for use by the KVM module > > > to indicate that a device is assigned a device driver can check that > > > flag and avoid destroying VFs while they are assigned and avoid system > > > failures. > > > > > > Signed-off-by: Greg Rose <gregory.v.rose@xxxxxxxxx> > > > --- > > > > > > include/linux/pci.h | 2 ++ > > > > I added Jesse and linux-pci to CC. > > > > > virt/kvm/assigned-dev.c | 2 ++ > > > virt/kvm/iommu.c | 4 ++++ > > > 3 files changed, 8 insertions(+), 0 deletions(-) > > > > I suppose this would also be useful in Xen's pciback or any other system > > which does passthrough? (Konrad CC'd for pciback) > > Definitely yes. Xen experiences the same issues when the PF driver is > removed while VFs are assigned to guests. > > > > > Is there some common lower layer we could hook this in to? (does > > iommu_attach/detach_device make sense?) Or shall we just add the flag > > manipulations to pciback as well? > [Greg Rose] > > I was unaware of any common lower layers, i.e I didn't know that Xen > also uses the iommu_attach/detach_device calls. my mistake -- under Xen the iommu is driver by the hypervisor and not the domain 0 kernel so there is no iommu_* in pciback. > It took me a week of digging around in the KVM module code just to > find these hooks. I'm not actually sure where in pciback the right place to put this would be is, perhaps Konrad has an idea. > Generally I stick to Ethernet device drivers and I'm not that > familiar with device pass through code. I was just confronted with a > problem and looking for some way to fix it. > > ;^) > > That sounds like a good idea, let me have a look at it. > > - Greg > -- Ian Campbell Traffic signals in New York are just rough guidelines. -- David Letterman -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
