Re: [PATCH net-next v2] net: mana: Add handler for sriov configure

Thank you for your contribution! Sashiko AI review found 2 potential issue(s) to consider:
- [High] Disabling SR-IOV without checking if Virtual Functions (VFs) are assigned to a guest VM.
- [High] Missing SR-IOV cleanup during PF driver removal and shutdown.
--

commit 3df823cc57367f69930deb6dc48172ce7d77e785
Author: Haiyang Zhang <[email protected]>

net: mana: Add handler for sriov configure

Add callback function for the pci_driver / sriov_configure.

It asks the NIC to provide certain number of VFs, or disable
VFs if the request is zero.

> diff --git a/drivers/net/ethernet/microsoft/mana/gdma_main.c b/drivers/net/ethernet/microsoft/mana/gdma_main.c
> index e8b7ffb47eb98..f7a79874fd0eb 100644
> --- a/drivers/net/ethernet/microsoft/mana/gdma_main.c
> +++ b/drivers/net/ethernet/microsoft/mana/gdma_main.c
> @@ -2496,6 +2496,20 @@ static void mana_gd_shutdown(struct pci_dev *pdev)
>  	pci_disable_device(pdev);
>  }

[Severity: High]
With the addition of SR-IOV support in this patch, do mana_gd_shutdown() and
mana_gd_remove() need to call pci_disable_sriov()?

If a user enables VFs via sysfs and then unbinds the driver or shuts down the
system, could this omission introduce a regression where orphaned, dead VF PCI
devices are left in the kernel?

Accessing these orphaned VFs might trigger use-after-free crashes or hardware
faults, as the underlying PF hardware and driver context have been destroyed.

> +static int mana_sriov_configure(struct pci_dev *pdev, int numvfs)
> +{
> +	int err = 0;
> +
> +	dev_info(&pdev->dev, "Requested num VFs: %d\n", numvfs);
> +
> +	if (numvfs > 0)
> +		err = pci_enable_sriov(pdev, numvfs);
> +	else
> +		pci_disable_sriov(pdev);

[Severity: High]
Should this path check if the VFs are currently assigned to a guest VM before
disabling SR-IOV?

If a privileged user writes 0 to sriov_numvfs while VFs are actively assigned
to a VM (e.g., via vfio-pci), could forcefully removing the PCI devices cause
a regression resulting in host IOMMU faults and guest kernel panics?

Would it be better to check pci_vfs_assigned(pdev) here, or use the
pci_sriov_configure_simple() helper to safely enforce this?

> +
> +	return err ? err : numvfs;
> +}

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/[email protected]?part=1