[PULL 12/18] vfio-user: IOMMU support for remote device

[Stefan Hajnoczi]

From: Jagannathan Raman <jag.raman@oracle.com>

Assign separate address space for each device in the remote processes.

Signed-off-by: Elena Ufimtseva <elena.ufimtseva@oracle.com>
Signed-off-by: John G Johnson <john.g.johnson@oracle.com>
Signed-off-by: Jagannathan Raman <jag.raman@oracle.com>
Reviewed-by: Stefan Hajnoczi <stefanha@redhat.com>
Message-id: 
afe0b0a97582cdad42b5b25636a29c523265a10a.1655151679.git.jag.raman@oracle.com
Signed-off-by: Stefan Hajnoczi <stefanha@redhat.com>
---
 MAINTAINERS               |   2 +
 include/hw/remote/iommu.h |  40 ++++++++++++
 hw/remote/iommu.c         | 131 ++++++++++++++++++++++++++++++++++++++
 hw/remote/machine.c       |  13 +++-
 hw/remote/meson.build     |   1 +
 5 files changed, 186 insertions(+), 1 deletion(-)
 create mode 100644 include/hw/remote/iommu.h
 create mode 100644 hw/remote/iommu.c

diff --git a/MAINTAINERS b/MAINTAINERS
index cbac72e239..563259101b 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -3644,6 +3644,8 @@ F: hw/remote/iohub.c
 F: include/hw/remote/iohub.h
 F: subprojects/libvfio-user
 F: hw/remote/vfio-user-obj.c
+F: hw/remote/iommu.c
+F: include/hw/remote/iommu.h
 
 EBPF:
 M: Jason Wang <jasowang@redhat.com>
diff --git a/include/hw/remote/iommu.h b/include/hw/remote/iommu.h
new file mode 100644
index 0000000000..33b68a8f4b
--- /dev/null
+++ b/include/hw/remote/iommu.h
@@ -0,0 +1,40 @@
+/**
+ * Copyright © 2022 Oracle and/or its affiliates.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#ifndef REMOTE_IOMMU_H
+#define REMOTE_IOMMU_H
+
+#include "hw/pci/pci_bus.h"
+#include "hw/pci/pci.h"
+
+#ifndef INT2VOIDP
+#define INT2VOIDP(i) (void *)(uintptr_t)(i)
+#endif
+
+typedef struct RemoteIommuElem {
+    MemoryRegion *mr;
+
+    AddressSpace as;
+} RemoteIommuElem;
+
+#define TYPE_REMOTE_IOMMU "x-remote-iommu"
+OBJECT_DECLARE_SIMPLE_TYPE(RemoteIommu, REMOTE_IOMMU)
+
+struct RemoteIommu {
+    Object parent;
+
+    GHashTable *elem_by_devfn;
+
+    QemuMutex lock;
+};
+
+void remote_iommu_setup(PCIBus *pci_bus);
+
+void remote_iommu_unplug_dev(PCIDevice *pci_dev);
+
+#endif
diff --git a/hw/remote/iommu.c b/hw/remote/iommu.c
new file mode 100644
index 0000000000..fd723d91f3
--- /dev/null
+++ b/hw/remote/iommu.c
@@ -0,0 +1,131 @@
+/**
+ * IOMMU for remote device
+ *
+ * Copyright © 2022 Oracle and/or its affiliates.
+ *
+ * This work is licensed under the terms of the GNU GPL, version 2 or later.
+ * See the COPYING file in the top-level directory.
+ *
+ */
+
+#include "qemu/osdep.h"
+
+#include "hw/remote/iommu.h"
+#include "hw/pci/pci_bus.h"
+#include "hw/pci/pci.h"
+#include "exec/memory.h"
+#include "exec/address-spaces.h"
+#include "trace.h"
+
+/**
+ * IOMMU for TYPE_REMOTE_MACHINE - manages DMA address space isolation
+ *     for remote machine. It is used by TYPE_VFIO_USER_SERVER.
+ *
+ * - Each TYPE_VFIO_USER_SERVER instance handles one PCIDevice on a PCIBus.
+ *   There is one RemoteIommu per PCIBus, so the RemoteIommu tracks multiple
+ *   PCIDevices by maintaining a ->elem_by_devfn mapping.
+ *
+ * - memory_region_init_iommu() is not used because vfio-user MemoryRegions
+ *   will be added to the elem->mr container instead. This is more natural
+ *   than implementing the IOMMUMemoryRegionClass APIs since vfio-user
+ *   provides something that is close to a full-fledged MemoryRegion and
+ *   not like an IOMMU mapping.
+ *
+ * - When a device is hot unplugged, the elem->mr reference is dropped so
+ *   all vfio-user MemoryRegions associated with this vfio-user server are
+ *   destroyed.
+ */
+
+static AddressSpace *remote_iommu_find_add_as(PCIBus *pci_bus,
+                                              void *opaque, int devfn)
+{
+    RemoteIommu *iommu = opaque;
+    RemoteIommuElem *elem = NULL;
+
+    qemu_mutex_lock(&iommu->lock);
+
+    elem = g_hash_table_lookup(iommu->elem_by_devfn, INT2VOIDP(devfn));
+
+    if (!elem) {
+        elem = g_malloc0(sizeof(RemoteIommuElem));
+        g_hash_table_insert(iommu->elem_by_devfn, INT2VOIDP(devfn), elem);
+    }
+
+    if (!elem->mr) {
+        elem->mr = MEMORY_REGION(object_new(TYPE_MEMORY_REGION));
+        memory_region_set_size(elem->mr, UINT64_MAX);
+        address_space_init(&elem->as, elem->mr, NULL);
+    }
+
+    qemu_mutex_unlock(&iommu->lock);
+
+    return &elem->as;
+}
+
+void remote_iommu_unplug_dev(PCIDevice *pci_dev)
+{
+    AddressSpace *as = pci_device_iommu_address_space(pci_dev);
+    RemoteIommuElem *elem = NULL;
+
+    if (as == &address_space_memory) {
+        return;
+    }
+
+    elem = container_of(as, RemoteIommuElem, as);
+
+    address_space_destroy(&elem->as);
+
+    object_unref(elem->mr);
+
+    elem->mr = NULL;
+}
+
+static void remote_iommu_init(Object *obj)
+{
+    RemoteIommu *iommu = REMOTE_IOMMU(obj);
+
+    iommu->elem_by_devfn = g_hash_table_new_full(NULL, NULL, NULL, g_free);
+
+    qemu_mutex_init(&iommu->lock);
+}
+
+static void remote_iommu_finalize(Object *obj)
+{
+    RemoteIommu *iommu = REMOTE_IOMMU(obj);
+
+    qemu_mutex_destroy(&iommu->lock);
+
+    g_hash_table_destroy(iommu->elem_by_devfn);
+
+    iommu->elem_by_devfn = NULL;
+}
+
+void remote_iommu_setup(PCIBus *pci_bus)
+{
+    RemoteIommu *iommu = NULL;
+
+    g_assert(pci_bus);
+
+    iommu = REMOTE_IOMMU(object_new(TYPE_REMOTE_IOMMU));
+
+    pci_setup_iommu(pci_bus, remote_iommu_find_add_as, iommu);
+
+    object_property_add_child(OBJECT(pci_bus), "remote-iommu", OBJECT(iommu));
+
+    object_unref(OBJECT(iommu));
+}
+
+static const TypeInfo remote_iommu_info = {
+    .name = TYPE_REMOTE_IOMMU,
+    .parent = TYPE_OBJECT,
+    .instance_size = sizeof(RemoteIommu),
+    .instance_init = remote_iommu_init,
+    .instance_finalize = remote_iommu_finalize,
+};
+
+static void remote_iommu_register_types(void)
+{
+    type_register_static(&remote_iommu_info);
+}
+
+type_init(remote_iommu_register_types)
diff --git a/hw/remote/machine.c b/hw/remote/machine.c
index 4d008ed721..cbb2add291 100644
--- a/hw/remote/machine.c
+++ b/hw/remote/machine.c
@@ -20,6 +20,7 @@
 #include "qapi/error.h"
 #include "hw/pci/pci_host.h"
 #include "hw/remote/iohub.h"
+#include "hw/remote/iommu.h"
 #include "hw/qdev-core.h"
 
 static void remote_machine_init(MachineState *machine)
@@ -99,6 +100,16 @@ static void remote_machine_instance_init(Object *obj)
     s->auto_shutdown = true;
 }
 
+static void remote_machine_dev_unplug_cb(HotplugHandler *hotplug_dev,
+                                         DeviceState *dev, Error **errp)
+{
+    qdev_unrealize(dev);
+
+    if (object_dynamic_cast(OBJECT(dev), TYPE_PCI_DEVICE)) {
+        remote_iommu_unplug_dev(PCI_DEVICE(dev));
+    }
+}
+
 static void remote_machine_class_init(ObjectClass *oc, void *data)
 {
     MachineClass *mc = MACHINE_CLASS(oc);
@@ -107,7 +118,7 @@ static void remote_machine_class_init(ObjectClass *oc, void 
*data)
     mc->init = remote_machine_init;
     mc->desc = "Experimental remote machine";
 
-    hc->unplug = qdev_simple_device_unplug_cb;
+    hc->unplug = remote_machine_dev_unplug_cb;
 
     object_class_property_add_bool(oc, "vfio-user",
                                    remote_machine_get_vfio_user,
diff --git a/hw/remote/meson.build b/hw/remote/meson.build
index 0eb5a0f375..ab25c04906 100644
--- a/hw/remote/meson.build
+++ b/hw/remote/meson.build
@@ -6,6 +6,7 @@ remote_ss.add(when: 'CONFIG_MULTIPROCESS', if_true: 
files('message.c'))
 remote_ss.add(when: 'CONFIG_MULTIPROCESS', if_true: files('remote-obj.c'))
 remote_ss.add(when: 'CONFIG_MULTIPROCESS', if_true: files('proxy.c'))
 remote_ss.add(when: 'CONFIG_MULTIPROCESS', if_true: files('iohub.c'))
+remote_ss.add(when: 'CONFIG_MULTIPROCESS', if_true: files('iommu.c'))
 remote_ss.add(when: 'CONFIG_VFIO_USER_SERVER', if_true: 
files('vfio-user-obj.c'))
 
 remote_ss.add(when: 'CONFIG_VFIO_USER_SERVER', if_true: libvfio_user_dep)
-- 
2.36.1