On 2016/5/25 5:11, Bjorn Helgaas wrote:

On Wed, Apr 27, 2016 at 08:43:27PM +0800, Yongji Xie wrote:

quoted

The capability of IRQ remapping is abstracted on IOMMU side on
some archs. There is a existing flag IOMMU_CAP_INTR_REMAP for this.

To have a universal flag to test this capability for different
archs on PCI side, we set PCI_BUS_FLAGS_MSI_REMAP for PCI buses
when IOMMU_CAP_INTR_REMAP is set.

Signed-off-by: Yongji Xie <redacted>
---
  drivers/iommu/iommu.c |   15 +++++++++++++++
  1 file changed, 15 insertions(+)

diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index 0e3b009..5d2b6f6 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c

@@ -813,6 +813,16 @@ struct iommu_group *pci_device_group(struct device *dev)
  	return group;
  }
  
+static void pci_check_msi_remapping(struct pci_dev *pdev,
+					const struct iommu_ops *ops)
+{
+	struct pci_bus *bus = pdev->bus;
+
+	if (ops->capable(IOMMU_CAP_INTR_REMAP) &&
+		!(bus->bus_flags & PCI_BUS_FLAGS_MSI_REMAP))
+		bus->bus_flags |= PCI_BUS_FLAGS_MSI_REMAP;
+}

This looks an awful lot like the pci_bus_check_msi_remapping() you add
elsewhere.  Why do we need both?

I will modify this function as you suggested.  And we add this function
here because some iommu drivers would be initialed after PCI probing.

quoted

  /**
   * iommu_group_get_for_dev - Find or create the IOMMU group for a device
   * @dev: target device

@@ -871,6 +881,9 @@ static int add_iommu_group(struct device *dev, void *data)
  	const struct iommu_ops *ops = cb->ops;
  	int ret;
  
+	if (dev_is_pci(dev) && ops->capable)
+		pci_check_msi_remapping(to_pci_dev(dev), ops);
+
  	if (!ops->add_device)
  		return 0;
  

@@ -913,6 +926,8 @@ static int iommu_bus_notifier(struct notifier_block *nb,
  	 * result in ADD/DEL notifiers to group->notifier
  	 */
  	if (action == BUS_NOTIFY_ADD_DEVICE) {
+		if (dev_is_pci(dev) && ops->capable)
+			pci_check_msi_remapping(to_pci_dev(dev), ops);

These calls don't smell right either.  Why do we need dev_is_pci()
checks here?

Some platform devices may also call this.

Can't this be done in the PCI probe path somehow, e.g.,
in pci_set_bus_msi_domain() or something?

Yes, this can be done in pci_create_root_bus(). But it could only
handle the case that iommu drivers are initialed before PCI probing.

Regards,
Yongji