[RFC,V3] virtio_pci: Add SR-IOV support

Hardware-realized virtio-pci devices can implement SR-IOV, so this
patch enables its use. The device in question is an upcoming Intel
NIC that implements both a virtio-net PF and virtio-net VFs. These
are hardware realizations of what has been up to now been a software
interface.

The device in question has the following 4-part PCI IDs:

PF: device: 1af4 vendor: 1041 subvendor: 8086 subdevice: 15fe
VF: device: 1af4 vendor: 1041 subvendor: 8086 subdevice: 05fe

The patch needs no check for device ID, because the callback will
never be made for devices that do not assert the capability or
when run on a platform incapable of SR-IOV.

One reason for this patch is because the hardware requires the
vendor ID of a VF to be the same as the vendor ID of the PF that
created it. So it seemed logical to simply have a fully-functioning
virtio-net PF create the VFs. This patch makes that possible.

Signed-off-by: Mark Rustad <mark.d.rustad@intel.com>
Reviewed-by: Alexander Duyck <alexander.h.duyck@intel.com>
---
Changes in V3:
- Move most code to a new helper in pci/iov.c, pci_sriov_configure
Changes in V2:
- Simplified logic from previous version, removed added driver variable
- Disable SR-IOV on driver removal excapt when VFs are assigned
- Sent as RFC to virtio-dev, linux-pci, netdev, lkml and others
---
 drivers/pci/iov.c                  |   48 ++++++++++++++++++++++++++++++++++++
 drivers/virtio/virtio_pci_common.c |    2 ++
 include/linux/pci.h                |   10 ++++++++
 3 files changed, 60 insertions(+)

From: Mark Rustad <mark.d.rustad@intel.com>
Date: Sun, 25 Feb 2018 19:19:11 -0800

> diff --git a/drivers/pci/iov.c b/drivers/pci/iov.c
> index 677924ae0350..ddd44a9d93ec 100644
> --- a/drivers/pci/iov.c
> +++ b/drivers/pci/iov.c
> @@ -367,6 +367,54 @@ static void sriov_disable(struct pci_dev *dev)
>  	pci_iov_set_numvfs(dev, 0);
>  }
>  
> +/**
> + * pci_sriov_disable - standard helper to disable SR-IOV
> + * @dev:the PCI PF device whose VFs are to be disabled
> + */
> +int pci_sriov_disable(struct pci_dev *dev)
> +{
> +	/*
> +	 * If vfs are assigned we cannot shut down SR-IOV without causing
> +	 * issues, so just leave the hardware available.
> +	 */
> +	if (pci_vfs_assigned(dev)) {
> +		pci_warn(&dev->dev,
> +			 "Cannot disable SR-IOV while VFs are assigned - VFs will not be deallocated\n");
> +		return -EPERM;
> +	}
> +	pci_disable_sriov(dev);
> +	return 0;
> +}
> +
> +static int pci_sriov_enable(struct pci_dev *dev, int num_vfs)
> +{
> +	int rc;
> +
> +	if (pci_num_vf(dev))
> +		return -EINVAL;
> +
> +	rc = pci_enable_sriov(dev, num_vfs);
> +	if (rc) {
> +		pci_warn(dev, "Failed to enable PCI sriov: %d\n", rc);
> +		return rc;
> +	}
> +	dev_info(dev, "SR-IOV enabled with %d VFs\n", num_vfs);
> +	return num_vfs;
> +}

I don't like these helpers on many different levels.

The pci_num_vf() test in pci_sriov_enable() is redundant, the pci_enable_sriov() code
path does that check and returns the same exact error code from sriov_enable().

Just call pci_enable_sriov() directly.  The log message adds no value justifying
an entirely new (and confusingly named) helper.  If the log message is useful, add
it to pci_enable_sriov().

Speaking of naming, is this stuff confusing or what?  As a programmer
what am I supposed to think when I consider what may be the difference
between two interfaces, the only difference in naming is that two
words are transposed?

	pci_enable_sriov()
	pci_sriov_enable()

	pci_disable_sriov()
	pci_sriov_disable()

?!?!?!?!

As per pci_sriov_disable() explicitly, all it does different is check
for vf assignment and return failure.

If you want a little help that does that, name it appropriately.

	pci_disable_sriov_if_unassigned()

So kill off pci_sriov_enable() helper completely, it is unnecessary,
and rename the disable helper so that it says something meaningful to
the reader.

Thanks.

> On Feb 27, 2018, at 7:35 AM, David Miller <davem@davemloft.net> wrote:
> 
> I don't like these helpers on many different levels.

<snip much embarrassment>

> So kill off pci_sriov_enable() helper completely, it is unnecessary,
> and rename the disable helper so that it says something meaningful to
> the reader.

Yes. Once pointed out, I completely agree with your comments and wish that I had seen those things myself.

> Thanks.

V3 was junk, but your comments apply to V4 as well, so please ignore it.

Thank you for your valuable review.