On Thu, Mar 23, 2017 at 12:29:28PM -0500, Bjorn Helgaas wrote:
> On Wed, Mar 01, 2017 at 03:31:58PM -0800, Himanshu Madhani wrote:
> > From: Michael Hernandez <michael.hernandez@xxxxxxxxxx>
> >
> > min_vecs is the minimum amount of vectors needed to operate in MSI-X mode
> > which may just include the vectors that don't need affinity.
> >
> > Disabling affinity settings causes the qla2xxx driver scsi_add_host
> > to fail when blk_mq is enabled as the blk_mq_pci_map_queues expects
> > affinity masks on each vector.
> >
> > v2 --> v3
> > o fixed code as per review comments.
> >
> > v1 --> v2
> >
> > o Moved the check from pci_alloc_irq_vectors_affinity() to
> > __pci_enable_{msi|msix}_range()
> >
> > Fixes: dfef358 ("PCI/MSI: Don't apply affinity if there aren't enough vectors left")
> > Signed-off-by: Michael Hernandez <michael.hernandez@xxxxxxxxxx>
> > Signed-off-by: Himanshu Madhani <himanshu.madhani@xxxxxxxxxx>
> > Cc: Christoph Hellwig <hch@xxxxxx>
> > Cc: stable@xxxxxxxxxxxxxxx
> > ---
> > drivers/pci/msi.c | 25 ++++++++++++++++++-------
> > 1 files changed, 18 insertions(+), 7 deletions(-)
> >
> > diff --git a/drivers/pci/msi.c b/drivers/pci/msi.c
> > index 7f73bac..46c0cdd 100644
> > --- a/drivers/pci/msi.c
> > +++ b/drivers/pci/msi.c
> > @@ -1091,6 +1091,15 @@ static int __pci_enable_msi_range(struct pci_dev *dev, int minvec, int maxvec,
> >
> > for (;;) {
> > if (affd) {
> > + /*
> > + * If there aren't any vectors left after applying the
> > + * pre/post vectors don't bother with assigning
> > + * affinity.
> > + */
> > + if (affd->pre_vectors + affd->post_vectors == nvec)
> > + affd = NULL;
> > + }
>
> I don't really like this because affd->pre_vectors and
> affd->post_vectors are not PCI MSI concepts. I think they really
> belong in irq/affinity.c, e.g., maybe this could be checked in
> irq_create_affinity_masks().
>
> If that could be done, we wouldn't have to duplicate the checks here
> for both MSI and MSI-X.
>
> I raised a similar question earlier:
> http://lkml.kernel.org/r/20170202173659.GD21267@xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx
I want to make progress on this, but that means a little conversation
here. Maybe we want to merge this patch as-is, but I'd like to hear
the reasons why it wouldn't really work to move the checks to
irq/affinity.c.
For now I'll mark it as "changes requested" in patchwork, which means
it will fall off my to-do list until it's reposted.
> > + if (affd) {
> > nvec = irq_calc_affinity_vectors(nvec, affd);
> > if (nvec < minvec)
> > return -ENOSPC;
> > @@ -1138,6 +1147,15 @@ static int __pci_enable_msix_range(struct pci_dev *dev,
> >
> > for (;;) {
> > if (affd) {
> > + /*
> > + * If there aren't any vectors left after applying the
> > + * pre/post vectors don't bother with assigning
> > + * affinity.
> > + */
> > + if (affd->pre_vectors + affd->post_vectors == nvec)
> > + affd = NULL;
> > + }
> > + if (affd) {
> > nvec = irq_calc_affinity_vectors(nvec, affd);
> > if (nvec < minvec)
> > return -ENOSPC;
> > @@ -1209,13 +1227,6 @@ int pci_alloc_irq_vectors_affinity(struct pci_dev *dev, unsigned int min_vecs,
> >
> > if (affd->pre_vectors + affd->post_vectors > min_vecs)
> > return -EINVAL;
> > -
> > - /*
> > - * If there aren't any vectors left after applying the pre/post
> > - * vectors don't bother with assigning affinity.
> > - */
> > - if (affd->pre_vectors + affd->post_vectors == min_vecs)
> > - affd = NULL;
> > } else {
> > if (WARN_ON(affd))
> > affd = NULL;
> > --
> > 1.7.1
> >
