On Tue, 2020-01-14 at 09:54 +0100, Christoph Hellwig wrote:
> On Fri, Jan 10, 2020 at 10:21:12AM -0700, Jon Derrick wrote:
> > Devices on the VMD domain use the VMD endpoint's requester ID and have
> > been relying on the VMD endpoint's DMA operations. The problem with this
> > was that VMD domain devices would use the VMD endpoint's attributes when
> > doing DMA and IOMMU mapping. We can be smarter about this by only using
> > the VMD endpoint when mapping and providing the correct child device's
> > attributes during DMA operations.
> >
> > This patch modifies Intel-IOMMU to check for a 'Direct DMA Alias' and
> > refer to it for mapping.
> >
> > Signed-off-by: Jon Derrick <jonathan.derrick@xxxxxxxxx>
> > ---
> > drivers/iommu/intel-iommu.c | 18 +++--
> > drivers/pci/controller/Kconfig | 1 -
> > drivers/pci/controller/vmd.c | 150 -----------------------------------------
> > 3 files changed, 13 insertions(+), 156 deletions(-)
> >
> > diff --git a/drivers/iommu/intel-iommu.c b/drivers/iommu/intel-iommu.c
> > index 716347e2..7ca807a 100644
> > --- a/drivers/iommu/intel-iommu.c
> > +++ b/drivers/iommu/intel-iommu.c
> > @@ -804,14 +804,14 @@ static struct intel_iommu *device_to_iommu(struct device *dev, u8 *bus, u8 *devf
> >
> > if (dev_is_pci(dev)) {
> > struct pci_dev *pf_pdev;
> > + struct pci_dev *dma_alias;
> >
> > pdev = to_pci_dev(dev);
> >
> > -#ifdef CONFIG_X86
> > - /* VMD child devices currently cannot be handled individually */
> > - if (is_vmd(pdev->bus))
> > - return NULL;
> > -#endif
>
> Don't we need this sanity check to prevent assingning vmd subdevices?
I don't think it's necessary now. The new code results in the child
devices being assigned the same IOMMU group as the VMD endpoint.
(AFAIK) You have to assign all devices in a group when doing
assignment, so by unbinding the VMD endpoint in order to assign it, you
would lose the child devices in the host.
>
> > + /* DMA aliased devices use the DMA alias's IOMMU */
> > + dma_alias = pci_direct_dma_alias(pdev);
> > + if (dma_alias)
> > + pdev = dma_alias;
> >
> > /* VFs aren't listed in scope tables; we need to look up
> > * the PF instead to find the IOMMU. */
> > @@ -2521,6 +2521,14 @@ struct dmar_domain *find_domain(struct device *dev)
> > dev->archdata.iommu == DUMMY_DEVICE_DOMAIN_INFO))
> > return NULL;
> >
> > + if (dev_is_pci(dev)) {
> > + struct pci_dev *pdev = to_pci_dev(dev);
> > + struct pci_dev *dma_alias = pci_direct_dma_alias(pdev);
> > +
> > + if (dma_alias)
> > + dev = &dma_alias->dev;
>
> Instead of all these duplicate calls, shouldn't pci_direct_dma_alias be
> replaced with a pci_real_dma_dev helper that either returns the
> dma parent if it exiѕts, or the actual device?
>
> Also I think this patch should be split - one for intel-iommu that
> just adds the real device checks, and then one that wires up vmd to
> the new mechanism and then removes all the cruft.
Thanks for the review. I'll work on the suggestion
