[PATCH v4 08/11] s390x/pci: use KVM-managed IOMMU for interpretation

[Matthew Rosato <mjrosato@xxxxxxxxxxxxx>]

When interpreting zPCI instructions, KVM will control the IOMMU mappings
in response to RPCIT instructions rather than relying on mapping ioctls
from userspace.  Mark the vfio device in pre_plug so that the appropriate
iommu domain will be allocated on the host during VFIO_SET_IOMMU.

Signed-off-by: Matthew Rosato <mjrosato@xxxxxxxxxxxxx>
---
 hw/s390x/s390-pci-bus.c          | 11 +++++++++++
 hw/s390x/s390-pci-vfio.c         | 22 ++++++++++++++++++++++
 include/hw/s390x/s390-pci-vfio.h |  5 +++++
 3 files changed, 38 insertions(+)

diff --git a/hw/s390x/s390-pci-bus.c b/hw/s390x/s390-pci-bus.c
index 5043b8c85c..513a276711 100644
--- a/hw/s390x/s390-pci-bus.c
+++ b/hw/s390x/s390-pci-bus.c
@@ -950,6 +950,17 @@ static void s390_pcihost_pre_plug(HotplugHandler *hotplug_dev, DeviceState *dev,
             error_setg(errp, "multifunction not supported in s390");
             return;
         }
+        /*
+         * If we have a vfio-pci device that wishes to use interpretation
+         * we must update the host IOMMU domain ops.
+         */
+        if (s390_pci_kvm_zpciop_allowed() &&
+            object_dynamic_cast(OBJECT(dev), "vfio-pci")) {
+            if (s390_pci_set_kvm_iommu(s, dev)) {
+                error_setg(errp, "KVM IOMMU not available for interpretation");
+                return;
+            }
+        }
     } else if (object_dynamic_cast(OBJECT(dev), TYPE_S390_PCI_DEVICE)) {
         S390PCIBusDevice *pbdev = S390_PCI_DEVICE(dev);
 
diff --git a/hw/s390x/s390-pci-vfio.c b/hw/s390x/s390-pci-vfio.c
index 4bf0a7e22d..7808e8d939 100644
--- a/hw/s390x/s390-pci-vfio.c
+++ b/hw/s390x/s390-pci-vfio.c
@@ -324,3 +324,25 @@ void s390_pci_get_clp_info(S390PCIBusDevice *pbdev)
 
     return;
 }
+
+/*
+ * This function will determine if the specified VFIOPCIDevice is linked to a
+ * zPCI device that requests interpretation support.  In this case, we must
+ * inform vfio that the KVM-managed IOMMU should be requested when the
+ * VFIO_SET_IOMMU ioctl is issued.
+ */
+int s390_pci_set_kvm_iommu(S390pciState *s, DeviceState *dev)
+{
+    VFIOPCIDevice *vdev = VFIO_PCI(dev);
+    S390PCIBusDevice *pbdev = s390_pci_find_dev_by_target(s, dev->id);
+
+    if (!pbdev) {
+        return -ENODEV;
+    }
+
+    if (pbdev->interp) {
+        vdev->kvm_managed_iommu = true;
+    }
+
+    return 0;
+}
diff --git a/include/hw/s390x/s390-pci-vfio.h b/include/hw/s390x/s390-pci-vfio.h
index 0c2e4b5175..5026f978c2 100644
--- a/include/hw/s390x/s390-pci-vfio.h
+++ b/include/hw/s390x/s390-pci-vfio.h
@@ -22,6 +22,7 @@ S390PCIDMACount *s390_pci_start_dma_count(S390pciState *s,
 void s390_pci_end_dma_count(S390pciState *s, S390PCIDMACount *cnt);
 bool s390_pci_get_host_fh(S390PCIBusDevice *pbdev, uint32_t *fh);
 void s390_pci_get_clp_info(S390PCIBusDevice *pbdev);
+int s390_pci_set_kvm_iommu(S390pciState *s, DeviceState *dev);
 #else
 static inline bool s390_pci_update_dma_avail(int fd, unsigned int *avail)
 {
@@ -40,6 +41,10 @@ static inline bool s390_pci_get_host_fh(S390PCIBusDevice *pbdev,
     return false;
 }
 static inline void s390_pci_get_clp_info(S390PCIBusDevice *pbdev) { }
+static inline int s390_pci_set_kvm_iommu(S390pciState *s, DeviceState *dev)
+{
+    return -EINVAL;
+}
 #endif
 
 #endif
-- 
2.27.0