Re: [PATCH] x86/pci: Some buggy virtual functions incorrectly report 1 for intx.

From: Alex Williamson
Date: Mon Jul 16 2018 - 17:18:02 EST


On Mon, 16 Jul 2018 13:42:25 -0700
Ashok Raj <ashok.raj@xxxxxxxxx> wrote:

> PCI_INTERRUPT_PIN should always read 0 for SRIOV Virtual Functions.
>
> Some SRIOV devices have some bugs in RTL and VF's end up reading 1
> instead of 0 for the PIN.
>
> We could enforce it by default in vfio_pci_nointx.
>
> Reported-by: Gage Eads <gage.eads@xxxxxxxxx>
> Tested-by: Gage Eads <gage.eads@xxxxxxxxx>
> Signed-off-by: Ashok Raj <ashok.raj@xxxxxxxxx>
> Cc: linux-kernel@xxxxxxxxxxxxxxx
> Cc: stable@xxxxxxxxxxxxxxx
> Cc: iommu@xxxxxxxxxxxxxxxxxxxxxxxxxx
> Cc: Joerg Roedel <joro@xxxxxxxxxx>
> Cc: Bjorn Helgaas <helgaas@xxxxxxxxxx>
> Cc: Gage Eads <gage.eads@xxxxxxxxx>
> ---
> drivers/vfio/pci/vfio_pci.c | 7 +++++++
> 1 file changed, 7 insertions(+)
>
> diff --git a/drivers/vfio/pci/vfio_pci.c b/drivers/vfio/pci/vfio_pci.c
> index b423a30..bc3f4fa 100644
> --- a/drivers/vfio/pci/vfio_pci.c
> +++ b/drivers/vfio/pci/vfio_pci.c
> @@ -192,6 +192,13 @@ static void vfio_pci_disable(struct vfio_pci_device *vdev);
> */
> static bool vfio_pci_nointx(struct pci_dev *pdev)
> {
> + /*
> + * Per PCI, no VF's should have INTx
> + * Simply disable it here
> + */
> + if (pdev->is_virtfn)
> + return true;
> +
> switch (pdev->vendor) {
> case PCI_VENDOR_ID_INTEL:
> switch (pdev->device) {

Nak, this is not what vfio_pci_nointx() is meant for, it's been tried
before and it was broken:

https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/drivers/vfio/pci/vfio_pci.c?id=834814e80268c818f354c8f402e0c6604ed75589

This would cause *every* VF to print:

dev_info(&pdev->dev, "Masking broken INTx support\n");

If the device is incapable of generating INTx then it's probably better
to drop it at vfio_pci_get_irq_count(). Please also note the subject is
misidentified, this is a "vfio/pci" patch, not an "x86/pci" patch.
kvm@vger is the official list for vfio patches. Thanks,

Alex