Re: [PATCH] nvme/pci: Provide SR-IOV support

[Christoph Hellwig <hch@infradead.org>]

On Thu, May 12, 2016 at 08:13:30AM -0600, Keith Busch wrote:
> This registers a standard boiler-plate sr-iov callback for nvme.

No real reason not to add it, but..


> +static int nvme_pci_sriov_configure(struct pci_dev *pdev, int numvfs)
> +{
> +	int ret = 0;
> +
> +	if (!pdev->is_physfn)
> +		return -EPERM;

pci_enable_sriov / pci_disable_sriov already check this.

> +	if (numvfs == 0) {
> +		if (pci_vfs_assigned(pdev)) {
> +			dev_warn(&pdev->dev,
> +				"Cannot disable SR-IOV VFs while assigned\n");
> +			return -EPERM;
> +		}

And pci_disable_sriov really should check this.

> +		pci_disable_sriov(pdev);
> +	} else
> +		ret = pci_enable_sriov(pdev, numvfs);

And these calls really should be done in common code.  At that point
there will be a whole lot less driver involvement with SR-IOV which
shouldn't exist, and we can split the callback into two for enable vs
disable instead of the overload that each driver then needs to
demultiplex.

I guess I can't just burden the second half on you, but at least
removing the first check and moving the second to the core would seem
like a good start.