Re: [PATCH v2] pci: fix device presence detection for VFs

[Michael S. Tsirkin]

On Wed, Nov 09, 2022 at 11:30:29AM -0600, Bjorn Helgaas wrote:
> On Wed, Nov 09, 2022 at 02:10:30AM -0500, Michael S. Tsirkin wrote:
> > On Tue, Nov 08, 2022 at 11:12:34PM -0600, Bjorn Helgaas wrote:
> > > On Wed, Nov 09, 2022 at 04:36:17AM +0000, Wei Gong wrote:
> > > > O Tue, Nov 08, 2022 at 01:02:35PM -0500, Michael S. Tsirkin wrote:
> > > > > On Tue, Nov 08, 2022 at 11:58:53AM -0600, Bjorn Helgaas wrote:
> > > > > > On Tue, Nov 08, 2022 at 10:19:07AM -0500, Michael S. Tsirkin wrote:
> > > > > > > On Tue, Nov 08, 2022 at 09:02:28AM -0600, Bjorn Helgaas wrote:
> > > > > > > > On Tue, Nov 08, 2022 at 08:53:00AM -0600, Bjorn Helgaas wrote:
> > > > > > > > > On Wed, Oct 26, 2022 at 02:11:21AM -0400, Michael S. Tsirkin wrote:
> > > > > > > > > > virtio uses the same driver for VFs and PFs.
> > > > > > > > > > Accordingly, pci_device_is_present is used to detect
> > > > > > > > > > device presence. This function isn't currently working
> > > > > > > > > > properly for VFs since it attempts reading device and
> > > > > > > > > > vendor ID.
> > > > > > > > > 
> > > > > > > > > > As VFs are present if and only if PF is present,
> > > > > > > > > > just return the value for that device.
> > > > > > > > > 
> > > > > > > > > VFs are only present when the PF is present *and* the PF
> > > > > > > > > has VF Enable set.  Do you care about the possibility that
> > > > > > > > > VF Enable has been cleared?
> > > > > > 
> > > > > > I think you missed this question.
> > > > > 
> > > > > I was hoping Wei will answer that, I don't have the hardware.
> > > > 
> > > > In my case I don't care that VF Enable has been cleared.
> > > 
> > > OK, let me rephrase that :)
> > > 
> > > I think pci_device_is_present(VF) should return "false" if the PF is
> > > present but VFs are disabled.
> > > 
> > > If you think it should return "true" when the PF is present and VFs
> > > are disabled, we should explain why.
> > > 
> > > We would also need to fix the commit log, because "VFs are present if
> > > and only if PF is present" is not actually true.  "VFs are present
> > > only if PF is present" is true, but "VFs are present if PF is present"
> > > is not.
> > 
> > Bjorn, I don't really understand the question.
> > 
> > How does one get a vf pointer without enabling sriov?
> > They are only created by sriov_add_vfs after calling
> > pcibios_sriov_enable.
> 
> Oh, I think I see where you're coming from.  The fact that we have a
> VF pointer means VFs were enabled in the past, and as long as the PF
> is still present, the VFs should still be enabled.
> 
> Since the continued existence of the VF device depends on VF Enable, I
> guess my question is whether we need to worry about VF Enable being
> cleared, e.g., via sysfs reset or a buggy PF driver.
> 
> Taking a step back, I don't understand the
> "if (!pci_device_is_present()) virtio_break_device()" strategy because
> checking for device presence is always unreliable.

The point is to break out of loops.


>  I assume the
> consumer of vq->broken, e.g., virtnet_send_command(), would see a
> failed PCI read that probably returns ~0 data.  Could it not check for
> that and then figure out whether that's valid data or an error
> indication?

No, it's not doing any reads - it is waiting for a DMA.

> It looks like today, virtnet_send_command() might sit in that "while"
> loop calling virtqueue_get_buf() repeatedly until virtio_pci_remove()
> notices the device is gone and marks it broken.  Something must be
> failing in virtqueue_get_buf() in that interval between the device
> disappearing and virtio_pci_remove() noticing it.
> 
> Bjorn

Nope - it is just doing posted writes, these disappear into thin ether
if there's no target.

-- 
MST