[Qemu-devel] [PATCH v4 09/12] spapr_iommu: add SPAPR VFIO IOMMU

[Alexey Kardashevskiy <aik@ozlabs.ru>]

This adds SPAPR VFIO IOMMU device in order to support DMA operations
for VFIO devices.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 hw/ppc/spapr_iommu.c   | 88 ++++++++++++++++++++++++++++++++++++++++++++++++++
 include/hw/ppc/spapr.h |  6 ++++
 2 files changed, 94 insertions(+)

diff --git a/hw/ppc/spapr_iommu.c b/hw/ppc/spapr_iommu.c
index 0b235b3..d91dee5 100644
--- a/hw/ppc/spapr_iommu.c
+++ b/hw/ppc/spapr_iommu.c
@@ -224,6 +224,76 @@ static target_ulong put_tce_emu(sPAPRTCETable *tcet, target_ulong ioba,
     return H_SUCCESS;
 }
 
+static IOMMUTLBEntry spapr_vfio_translate_iommu(MemoryRegion *iommu,
+                                                hwaddr addr)
+{
+    IOMMUTLBEntry entry;
+    /*
+     * This callback would normally be used by a QEMU device for DMA
+     * but in this case the vfio-pci device does not do any DMA.
+     * Instead, the real hardware does DMA and hardware TCE table
+     * performs the address translation.
+     */
+    assert(0);
+    return entry;
+}
+
+static MemoryRegionIOMMUOps spapr_vfio_iommu_ops = {
+    .translate = spapr_vfio_translate_iommu,
+};
+
+static int spapr_tce_table_vfio_realize(DeviceState *dev)
+{
+    sPAPRTCETable *tcet = SPAPR_TCE_TABLE(dev);
+
+    memory_region_init_iommu(&tcet->iommu, NULL, &spapr_vfio_iommu_ops,
+                             "iommu-vfio-spapr", UINT64_MAX);
+
+    QLIST_INSERT_HEAD(&spapr_tce_tables, tcet, list);
+
+    return 0;
+}
+
+sPAPRTCETable *spapr_vfio_new_table(DeviceState *owner, uint32_t liobn,
+                                    int group_fd)
+{
+    sPAPRTCETable *tcet;
+
+    if (spapr_tce_find_by_liobn(liobn)) {
+        fprintf(stderr, "Attempted to create TCE table with duplicate"
+                " LIOBN 0x%x\n", liobn);
+        return NULL;
+    }
+
+    tcet = SPAPR_TCE_TABLE(object_new(TYPE_SPAPR_TCE_TABLE_VFIO));
+    tcet->liobn = liobn;
+    object_property_add_child(OBJECT(owner), "tce-table", OBJECT(tcet), NULL);
+
+    object_property_set_bool(OBJECT(tcet), true, "realized", NULL);
+
+    return tcet;
+}
+
+static target_ulong put_tce_vfio(sPAPRTCETable *tcet, target_ulong ioba,
+                                 target_ulong tce)
+{
+    IOMMUTLBEntry entry;
+
+    entry.iova = ioba & ~SPAPR_TCE_PAGE_MASK;
+    entry.translated_addr = tce & ~SPAPR_TCE_PAGE_MASK;
+    entry.addr_mask = SPAPR_TCE_PAGE_MASK;
+    entry.perm = 0;
+    if ((tce & SPAPR_TCE_RO) == SPAPR_TCE_RO) {
+        entry.perm |= IOMMU_RO;
+    }
+    if ((tce & SPAPR_TCE_WO) == SPAPR_TCE_WO) {
+        entry.perm |= IOMMU_WO;
+    }
+    memory_region_notify_iommu(&tcet->iommu, entry);
+
+    return H_SUCCESS;
+}
+
 static target_ulong h_put_tce_indirect(PowerPCCPU *cpu,
                                        sPAPREnvironment *spapr,
                                        target_ulong opcode, target_ulong *args)
@@ -406,9 +476,27 @@ static TypeInfo spapr_tce_table_info = {
     .instance_finalize = spapr_tce_table_finalize,
 };
 
+static void spapr_tce_table_vfio_class_init(ObjectClass *klass, void *data)
+{
+    DeviceClass *dc = DEVICE_CLASS(klass);
+    sPAPRTCETableClass *stc = SPAPR_TCE_TABLE_CLASS(klass);
+
+    dc->init = spapr_tce_table_vfio_realize;
+    stc->put_tce = put_tce_vfio;
+}
+
+static TypeInfo spapr_tce_table_vfio_info = {
+    .name = TYPE_SPAPR_TCE_TABLE_VFIO,
+    .parent = TYPE_SPAPR_TCE_TABLE,
+    .instance_size = sizeof(sPAPRTCETable),
+    .class_init = spapr_tce_table_vfio_class_init,
+    .class_size = sizeof(sPAPRTCETableClass),
+};
+
 static void register_types(void)
 {
     type_register_static(&spapr_tce_table_info);
+    type_register_static(&spapr_tce_table_vfio_info);
 }
 
 type_init(register_types);
diff --git a/include/hw/ppc/spapr.h b/include/hw/ppc/spapr.h
index a54de97..7054bdb 100644
--- a/include/hw/ppc/spapr.h
+++ b/include/hw/ppc/spapr.h
@@ -369,6 +369,10 @@ typedef struct sPAPRTCETable sPAPRTCETable;
 #define SPAPR_TCE_TABLE(obj) \
     OBJECT_CHECK(sPAPRTCETable, (obj), TYPE_SPAPR_TCE_TABLE)
 
+#define TYPE_SPAPR_TCE_TABLE_VFIO "spapr-tce-table-vfio"
+#define SPAPR_TCE_TABLE_VFIO(obj) \
+    OBJECT_CHECK(sPAPRTCETable, (obj), TYPE_SPAPR_TCE_TABLE_VFIO)
+
 #define SPAPR_TCE_TABLE_CLASS(klass) \
      OBJECT_CLASS_CHECK(sPAPRTCETableClass, (klass), TYPE_SPAPR_TCE_TABLE)
 #define SPAPR_TCE_TABLE_GET_CLASS(obj) \
@@ -397,6 +401,8 @@ void spapr_events_init(sPAPREnvironment *spapr);
 void spapr_events_fdt_skel(void *fdt, uint32_t epow_irq);
 sPAPRTCETable *spapr_tce_new_table(DeviceState *owner, uint32_t liobn,
                                    size_t window_size);
+sPAPRTCETable *spapr_vfio_new_table(DeviceState *owner, uint32_t liobn,
+                                    int group_fd);
 MemoryRegion *spapr_tce_get_iommu(sPAPRTCETable *tcet);
 void spapr_tce_set_bypass(sPAPRTCETable *tcet, bool bypass);
 int spapr_dma_dt(void *fdt, int node_off, const char *propname,
-- 
1.8.4.rc4