Re: [stable][PATCH] PCIe hot-plug for Intel IOMMU

[David Woodhouse <dwmw2@infradead.org>]

On Wed, 2009-11-11 at 13:27 -0800, Yinghai Lu wrote:
> On Wed, Nov 11, 2009 at 7:23 AM, Fenghua Yu <fenghua.yu@intel.com> wrote:
> > To support PCIe hot plug in IOMMU, we register a notifier to respond to device
> > change action.
> >
> > When the notifier gets BUS_NOTIFY_UNBOUND_DRIVER, it removes the device from its
> > DMAR domain.
> >
> > A hot added device will be added into an IOMMU domain when it first does IOMMU
> > op. So there is no need to add more code for hot add.
> >
> > Without the patch, after a hot-remove, a hot-added device on the same slot will
> > not work.
> >
> > Signed-off-by: Fenghua Yu <fenghua.yu@intel.com>
> 
> Tested-by: Yinghai Lu <yinghai@kernel.org>

But not with 'intel_iommu=igfx_off', I note. The call to find_domain()
will oops when it runs on the graphics device with ->archdata.iommu ==
DUMMY_DEVICE_DOMAIN_INFO (== -1). All other calls to find_domain() are
guaranteed not to happen for such devices.

One might argue that find_domain() should cope, but do we really want to
be doing that extra check in the unmap fast path?

Testing this fix now...

diff --git a/drivers/pci/intel-iommu.c b/drivers/pci/intel-iommu.c
index 1840a05..f44a015 100644
--- a/drivers/pci/intel-iommu.c
+++ b/drivers/pci/intel-iommu.c
@@ -3228,6 +3228,9 @@ static int device_notifier(struct notifier_block *nb,
 	struct pci_dev *pdev = to_pci_dev(dev);
 	struct dmar_domain *domain;
 
+	if (iommu_no_mapping(dev))
+		return 0;
+
 	domain = find_domain(pdev);
 	if (!domain)
 		return 0;


-- 
David Woodhouse                            Open Source Technology Centre
David.Woodhouse@intel.com                              Intel Corporation