On Fri, 29 Jun 2018 15:09:34 -0500, Bjorn Helgaas wrote:
> On Mon, Jun 18, 2018 at 03:01:42PM -0700, Jakub Kicinski wrote:
> > Bjorn points out that currently core and most of the drivers don't
> > clean up dev->sriov->driver_max_VFs settings on .remove(). This
> > means that if a different driver is bound afterwards it will
> > inherit the old setting:
> >
> > - load PF driver 1
> > - driver 1 calls pci_sriov_set_totalvfs() to reduce driver_max_VFs
> > - unload PF driver 1
> > - load PF driver 2
> > # driver 2 does not know to call pci_sriov_set_totalvfs()
> >
> > Some drivers (e.g. nfp) used to do the clean up by calling
> > pci_sriov_set_totalvfs(dev, 0), since 0 was equivalent to no driver
> > limit set. After commit 8d85a7a4f2c9 ("PCI/IOV: Allow PF drivers
> > to limit total_VFs to 0") 0 no longer has this special meaning.
> >
> > The need to reset driver_max_VFs comes from the fact that old FW
> > builds may not expose its VF limits to the drivers, and depend on
> > the ability to reject the configuration change when driver notifies
> > the FW as part of struct pci_driver->sriov_configure() callback.
> > Therefore if there is no information on VF count limits we should
> > use the PCI capability max, and not the last value set by
> > pci_sriov_set_totalvfs().
> >
> > Reset driver_max_VFs back to total_VFs after device remove. If
> > drivers want to reload FW/reconfigure the device while driver is
> > bound we may add an explicit pci_sriov_reset_totalvfs(), but for
> > now no driver is doing that.
> >
> > Fixes: 8d85a7a4f2c9 ("PCI/IOV: Allow PF drivers to limit total_VFs to 0")
> > Signed-off-by: Jakub Kicinski <jakub.kicinski@xxxxxxxxxxxxx>
>
> Applied to for-linus for v4.18, thanks!
Awesome, thanks a lot!
