Re: [PATCH] PCI: IOV: reread SRIOV_NUM_VF after enabling ARI

[Bjorn Helgaas <helgaas@kernel.org>]

On Wed, Oct 07, 2015 at 03:21:36PM -0500, Ben Shelton wrote:
> Hi Bjorn,
> 
> On Wed, Oct 07, 2015 at 02:29:40PM -0500, Bjorn Helgaas wrote:
> > Hi Ben,
> > 
> > On Fri, Sep 11, 2015 at 04:55:00PM -0500, Ben Shelton wrote:
> > > For some SR-IOV devices, the number of available virtual functions
> > > increases after enabling ARI.  Currently, SRIOV_NUM_VF is read and saved
> > > off before the ARI control bit is enabled.  This causes an issue when VFs
> > > are enabled.
> > > 
> > > At device init, SRIOV_INITIAL_VF and SRIOV_NUM_VF are specified to contain
> > > the number of available VFs for the device.  sriov_enable() does a sanity
> > > check that PCI_SRIOV_INITIAL_VF is not greater than iov->total_VFs, the
> > > saved-off value of SRIOV_NUM_VF.  Since the value of both SRIOV_INITIAL_VF
> > > and SRIOV_NUM_VF has increased after enabling the ARI bit, the check fails,
> > > and the VFs cannot be enabled.
> > > 
> > > To fix the issue, after ARI is enabled for a device, reread SRIOV_NUM_VF.
> > > 
> > > Signed-off-by: Ben Shelton <benjamin.h.shelton@intel.com>
> > > ---
> > >  drivers/pci/iov.c | 9 ++++++++-
> > >  1 file changed, 8 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/pci/iov.c b/drivers/pci/iov.c
> > > index ee0ebff..88b1959 100644
> > > --- a/drivers/pci/iov.c
> > > +++ b/drivers/pci/iov.c
> > > @@ -388,6 +388,7 @@ static int sriov_init(struct pci_dev *dev, int pos)
> > >  	struct pci_sriov *iov;
> > >  	struct resource *res;
> > >  	struct pci_dev *pdev;
> > > +	bool total_needs_reread = false;
> > >  
> > >  	if (pci_pcie_type(dev) != PCI_EXP_TYPE_RC_END &&
> > >  	    pci_pcie_type(dev) != PCI_EXP_TYPE_ENDPOINT)
> > > @@ -409,12 +410,18 @@ static int sriov_init(struct pci_dev *dev, int pos)
> > >  			goto found;
> > >  
> > >  	pdev = NULL;
> > > -	if (pci_ari_enabled(dev->bus))
> > > +	if (pci_ari_enabled(dev->bus)) {
> > >  		ctrl |= PCI_SRIOV_CTRL_ARI;
> > > +		total_needs_reread = true;
> > > +	}
> > >  
> > >  found:
> > >  	pci_write_config_word(dev, pos + PCI_SRIOV_CTRL, ctrl);
> > >  	pci_write_config_word(dev, pos + PCI_SRIOV_NUM_VF, 0);
> > > +
> > > +	if (total_needs_reread)
> > > +		pci_read_config_word(dev, pos + PCI_SRIOV_TOTAL_VF, &total);
> > 
> > Can we just *move* the PCI_SRIOV_TOTAL_VF read from its original
> > location and do it unconditionally here?  I don't think we use "total"
> > for anything in the interim, and it'd be nice if we could do this
> > without the "total_needs_reread" flag.
> 
> The only thing we do with 'total' is return early if it's read as 0
> (i.e. for a device that does not support any VFs).  If we just move the
> read to here, a PCI device with no VFs would have its PCI_SRIOV_CTRL
> config word written before PCI_SRIOV_TOTAL_VF is read as 0 and
> sriov_init() returns.  The reread preserves the original behavior.
> 
> If the change in behavior is OK, I'll make the change and submit a v2.

I think that's OK.  We do return some other errors after modifying
PCI_SRIOV_CTRL, so I don't know that this would be any worse.

Bjorn