KVM: arm64: vgic-its: Implement ITS command queue command handlers
[linux-2.6-block.git] / virt / kvm / arm / vgic / vgic-its.c
index 6f43b3b1172b73dcbc2d3b0ed5ace3ba01bf9367..1408c88d063e17ba64508c6484e7691cd5994d2f 100644 (file)
 #include "vgic.h"
 #include "vgic-mmio.h"
 
+/*
+ * Creates a new (reference to a) struct vgic_irq for a given LPI.
+ * If this LPI is already mapped on another ITS, we increase its refcount
+ * and return a pointer to the existing structure.
+ * If this is a "new" LPI, we allocate and initialize a new struct vgic_irq.
+ * This function returns a pointer to the _unlocked_ structure.
+ */
+static struct vgic_irq *vgic_add_lpi(struct kvm *kvm, u32 intid)
+{
+       struct vgic_dist *dist = &kvm->arch.vgic;
+       struct vgic_irq *irq = vgic_get_irq(kvm, NULL, intid), *oldirq;
+
+       /* In this case there is no put, since we keep the reference. */
+       if (irq)
+               return irq;
+
+       irq = kzalloc(sizeof(struct vgic_irq), GFP_KERNEL);
+       if (!irq)
+               return NULL;
+
+       INIT_LIST_HEAD(&irq->lpi_list);
+       INIT_LIST_HEAD(&irq->ap_list);
+       spin_lock_init(&irq->irq_lock);
+
+       irq->config = VGIC_CONFIG_EDGE;
+       kref_init(&irq->refcount);
+       irq->intid = intid;
+
+       spin_lock(&dist->lpi_list_lock);
+
+       /*
+        * There could be a race with another vgic_add_lpi(), so we need to
+        * check that we don't add a second list entry with the same LPI.
+        */
+       list_for_each_entry(oldirq, &dist->lpi_list_head, lpi_list) {
+               if (oldirq->intid != intid)
+                       continue;
+
+               /* Someone was faster with adding this LPI, lets use that. */
+               kfree(irq);
+               irq = oldirq;
+
+               /*
+                * This increases the refcount, the caller is expected to
+                * call vgic_put_irq() on the returned pointer once it's
+                * finished with the IRQ.
+                */
+               kref_get(&irq->refcount);
+
+               goto out_unlock;
+       }
+
+       list_add_tail(&irq->lpi_list, &dist->lpi_list_head);
+       dist->lpi_list_count++;
+
+out_unlock:
+       spin_unlock(&dist->lpi_list_lock);
+
+       return irq;
+}
+
 struct its_device {
        struct list_head dev_list;
 
@@ -62,16 +123,75 @@ struct its_itte {
        u32 event_id;
 };
 
+/*
+ * Find and returns a device in the device table for an ITS.
+ * Must be called with the its_lock mutex held.
+ */
+static struct its_device *find_its_device(struct vgic_its *its, u32 device_id)
+{
+       struct its_device *device;
+
+       list_for_each_entry(device, &its->device_list, dev_list)
+               if (device_id == device->device_id)
+                       return device;
+
+       return NULL;
+}
+
+/*
+ * Find and returns an interrupt translation table entry (ITTE) for a given
+ * Device ID/Event ID pair on an ITS.
+ * Must be called with the its_lock mutex held.
+ */
+static struct its_itte *find_itte(struct vgic_its *its, u32 device_id,
+                                 u32 event_id)
+{
+       struct its_device *device;
+       struct its_itte *itte;
+
+       device = find_its_device(its, device_id);
+       if (device == NULL)
+               return NULL;
+
+       list_for_each_entry(itte, &device->itt_head, itte_list)
+               if (itte->event_id == event_id)
+                       return itte;
+
+       return NULL;
+}
+
+/* To be used as an iterator this macro misses the enclosing parentheses */
+#define for_each_lpi_its(dev, itte, its) \
+       list_for_each_entry(dev, &(its)->device_list, dev_list) \
+               list_for_each_entry(itte, &(dev)->itt_head, itte_list)
+
 /*
  * We only implement 48 bits of PA at the moment, although the ITS
  * supports more. Let's be restrictive here.
  */
+#define BASER_ADDRESS(x)       ((x) & GENMASK_ULL(47, 16))
 #define CBASER_ADDRESS(x)      ((x) & GENMASK_ULL(47, 12))
 #define PENDBASER_ADDRESS(x)   ((x) & GENMASK_ULL(47, 16))
 #define PROPBASER_ADDRESS(x)   ((x) & GENMASK_ULL(47, 12))
 
 #define GIC_LPI_OFFSET 8192
 
+/*
+ * Finds and returns a collection in the ITS collection table.
+ * Must be called with the its_lock mutex held.
+ */
+static struct its_collection *find_collection(struct vgic_its *its, int coll_id)
+{
+       struct its_collection *collection;
+
+       list_for_each_entry(collection, &its->collection_list, coll_list) {
+               if (coll_id == collection->collection_id)
+                       return collection;
+       }
+
+       return NULL;
+}
+
 #define LPI_PROP_ENABLE_BIT(p) ((p) & LPI_PROP_ENABLED)
 #define LPI_PROP_PRIORITY(p)   ((p) & 0xfc)
 
@@ -144,6 +264,51 @@ static int vgic_copy_lpi_list(struct kvm *kvm, u32 **intid_ptr)
        return irq_count;
 }
 
+/*
+ * Promotes the ITS view of affinity of an ITTE (which redistributor this LPI
+ * is targeting) to the VGIC's view, which deals with target VCPUs.
+ * Needs to be called whenever either the collection for a LPIs has
+ * changed or the collection itself got retargeted.
+ */
+static void update_affinity_itte(struct kvm *kvm, struct its_itte *itte)
+{
+       struct kvm_vcpu *vcpu;
+
+       if (!its_is_collection_mapped(itte->collection))
+               return;
+
+       vcpu = kvm_get_vcpu(kvm, itte->collection->target_addr);
+
+       spin_lock(&itte->irq->irq_lock);
+       itte->irq->target_vcpu = vcpu;
+       spin_unlock(&itte->irq->irq_lock);
+}
+
+/*
+ * Updates the target VCPU for every LPI targeting this collection.
+ * Must be called with the its_lock mutex held.
+ */
+static void update_affinity_collection(struct kvm *kvm, struct vgic_its *its,
+                                      struct its_collection *coll)
+{
+       struct its_device *device;
+       struct its_itte *itte;
+
+       for_each_lpi_its(device, itte, its) {
+               if (!itte->collection || coll != itte->collection)
+                       continue;
+
+               update_affinity_itte(kvm, itte);
+       }
+}
+
+static u32 max_lpis_propbaser(u64 propbaser)
+{
+       int nr_idbits = (propbaser & 0x1f) + 1;
+
+       return 1U << min(nr_idbits, INTERRUPT_ID_BITS_ITS);
+}
+
 /*
  * Scan the whole LPI pending table and sync the pending bit in there
  * with our own data structures. This relies on the LPI being
@@ -283,10 +448,504 @@ static void its_free_itte(struct kvm *kvm, struct its_itte *itte)
        kfree(itte);
 }
 
+static u64 its_cmd_mask_field(u64 *its_cmd, int word, int shift, int size)
+{
+       return (le64_to_cpu(its_cmd[word]) >> shift) & (BIT_ULL(size) - 1);
+}
+
+#define its_cmd_get_command(cmd)       its_cmd_mask_field(cmd, 0,  0,  8)
+#define its_cmd_get_deviceid(cmd)      its_cmd_mask_field(cmd, 0, 32, 32)
+#define its_cmd_get_id(cmd)            its_cmd_mask_field(cmd, 1,  0, 32)
+#define its_cmd_get_physical_id(cmd)   its_cmd_mask_field(cmd, 1, 32, 32)
+#define its_cmd_get_collection(cmd)    its_cmd_mask_field(cmd, 2,  0, 16)
+#define its_cmd_get_target_addr(cmd)   its_cmd_mask_field(cmd, 2, 16, 32)
+#define its_cmd_get_validbit(cmd)      its_cmd_mask_field(cmd, 2, 63,  1)
+
+/*
+ * The DISCARD command frees an Interrupt Translation Table Entry (ITTE).
+ * Must be called with the its_lock mutex held.
+ */
+static int vgic_its_cmd_handle_discard(struct kvm *kvm, struct vgic_its *its,
+                                      u64 *its_cmd)
+{
+       u32 device_id = its_cmd_get_deviceid(its_cmd);
+       u32 event_id = its_cmd_get_id(its_cmd);
+       struct its_itte *itte;
+
+
+       itte = find_itte(its, device_id, event_id);
+       if (itte && itte->collection) {
+               /*
+                * Though the spec talks about removing the pending state, we
+                * don't bother here since we clear the ITTE anyway and the
+                * pending state is a property of the ITTE struct.
+                */
+               its_free_itte(kvm, itte);
+               return 0;
+       }
+
+       return E_ITS_DISCARD_UNMAPPED_INTERRUPT;
+}
+
+/*
+ * The MOVI command moves an ITTE to a different collection.
+ * Must be called with the its_lock mutex held.
+ */
+static int vgic_its_cmd_handle_movi(struct kvm *kvm, struct vgic_its *its,
+                                   u64 *its_cmd)
+{
+       u32 device_id = its_cmd_get_deviceid(its_cmd);
+       u32 event_id = its_cmd_get_id(its_cmd);
+       u32 coll_id = its_cmd_get_collection(its_cmd);
+       struct kvm_vcpu *vcpu;
+       struct its_itte *itte;
+       struct its_collection *collection;
+
+       itte = find_itte(its, device_id, event_id);
+       if (!itte)
+               return E_ITS_MOVI_UNMAPPED_INTERRUPT;
+
+       if (!its_is_collection_mapped(itte->collection))
+               return E_ITS_MOVI_UNMAPPED_COLLECTION;
+
+       collection = find_collection(its, coll_id);
+       if (!its_is_collection_mapped(collection))
+               return E_ITS_MOVI_UNMAPPED_COLLECTION;
+
+       itte->collection = collection;
+       vcpu = kvm_get_vcpu(kvm, collection->target_addr);
+
+       spin_lock(&itte->irq->irq_lock);
+       itte->irq->target_vcpu = vcpu;
+       spin_unlock(&itte->irq->irq_lock);
+
+       return 0;
+}
+
+static void vgic_its_init_collection(struct vgic_its *its,
+                                    struct its_collection *collection,
+                                    u32 coll_id)
+{
+       collection->collection_id = coll_id;
+       collection->target_addr = COLLECTION_NOT_MAPPED;
+
+       list_add_tail(&collection->coll_list, &its->collection_list);
+}
+
+/*
+ * The MAPTI and MAPI commands map LPIs to ITTEs.
+ * Must be called with its_lock mutex held.
+ */
+static int vgic_its_cmd_handle_mapi(struct kvm *kvm, struct vgic_its *its,
+                                   u64 *its_cmd, u8 subcmd)
+{
+       u32 device_id = its_cmd_get_deviceid(its_cmd);
+       u32 event_id = its_cmd_get_id(its_cmd);
+       u32 coll_id = its_cmd_get_collection(its_cmd);
+       struct its_itte *itte;
+       struct its_device *device;
+       struct its_collection *collection, *new_coll = NULL;
+       int lpi_nr;
+
+       device = find_its_device(its, device_id);
+       if (!device)
+               return E_ITS_MAPTI_UNMAPPED_DEVICE;
+
+       collection = find_collection(its, coll_id);
+       if (!collection) {
+               new_coll = kzalloc(sizeof(struct its_collection), GFP_KERNEL);
+               if (!new_coll)
+                       return -ENOMEM;
+       }
+
+       if (subcmd == GITS_CMD_MAPTI)
+               lpi_nr = its_cmd_get_physical_id(its_cmd);
+       else
+               lpi_nr = event_id;
+       if (lpi_nr < GIC_LPI_OFFSET ||
+           lpi_nr >= max_lpis_propbaser(kvm->arch.vgic.propbaser)) {
+               kfree(new_coll);
+               return E_ITS_MAPTI_PHYSICALID_OOR;
+       }
+
+       itte = find_itte(its, device_id, event_id);
+       if (!itte) {
+               itte = kzalloc(sizeof(struct its_itte), GFP_KERNEL);
+               if (!itte) {
+                       kfree(new_coll);
+                       return -ENOMEM;
+               }
+
+               itte->event_id  = event_id;
+               list_add_tail(&itte->itte_list, &device->itt_head);
+       }
+
+       if (!collection) {
+               collection = new_coll;
+               vgic_its_init_collection(its, collection, coll_id);
+       }
+
+       itte->collection = collection;
+       itte->lpi = lpi_nr;
+       itte->irq = vgic_add_lpi(kvm, lpi_nr);
+       update_affinity_itte(kvm, itte);
+
+       /*
+        * We "cache" the configuration table entries in out struct vgic_irq's.
+        * However we only have those structs for mapped IRQs, so we read in
+        * the respective config data from memory here upon mapping the LPI.
+        */
+       update_lpi_config(kvm, itte->irq, NULL);
+
+       return 0;
+}
+
+/* Requires the its_lock to be held. */
+static void vgic_its_unmap_device(struct kvm *kvm, struct its_device *device)
+{
+       struct its_itte *itte, *temp;
+
+       /*
+        * The spec says that unmapping a device with still valid
+        * ITTEs associated is UNPREDICTABLE. We remove all ITTEs,
+        * since we cannot leave the memory unreferenced.
+        */
+       list_for_each_entry_safe(itte, temp, &device->itt_head, itte_list)
+               its_free_itte(kvm, itte);
+
+       list_del(&device->dev_list);
+       kfree(device);
+}
+
+/*
+ * Check whether a device ID can be stored into the guest device tables.
+ * For a direct table this is pretty easy, but gets a bit nasty for
+ * indirect tables. We check whether the resulting guest physical address
+ * is actually valid (covered by a memslot and guest accessbible).
+ * For this we have to read the respective first level entry.
+ */
+static bool vgic_its_check_device_id(struct kvm *kvm, struct vgic_its *its,
+                                    int device_id)
+{
+       u64 r = its->baser_device_table;
+       int nr_entries = GITS_BASER_NR_PAGES(r) * SZ_64K;
+       int index;
+       u64 indirect_ptr;
+       gfn_t gfn;
+
+
+       if (!(r & GITS_BASER_INDIRECT))
+               return device_id < (nr_entries / GITS_BASER_ENTRY_SIZE(r));
+
+       /* calculate and check the index into the 1st level */
+       index = device_id / (SZ_64K / GITS_BASER_ENTRY_SIZE(r));
+       if (index >= (nr_entries / sizeof(u64)))
+               return false;
+
+       /* Each 1st level entry is represented by a 64-bit value. */
+       if (!kvm_read_guest(kvm,
+                           BASER_ADDRESS(r) + index * sizeof(indirect_ptr),
+                           &indirect_ptr, sizeof(indirect_ptr)))
+               return false;
+
+       /* check the valid bit of the first level entry */
+       if (!(indirect_ptr & BIT_ULL(63)))
+               return false;
+
+       /*
+        * Mask the guest physical address and calculate the frame number.
+        * Any address beyond our supported 48 bits of PA will be caught
+        * by the actual check in the final step.
+        */
+       gfn = (indirect_ptr & GENMASK_ULL(51, 16)) >> PAGE_SHIFT;
+
+       return kvm_is_visible_gfn(kvm, gfn);
+}
+
+/*
+ * MAPD maps or unmaps a device ID to Interrupt Translation Tables (ITTs).
+ * Must be called with the its_lock mutex held.
+ */
+static int vgic_its_cmd_handle_mapd(struct kvm *kvm, struct vgic_its *its,
+                                   u64 *its_cmd)
+{
+       u32 device_id = its_cmd_get_deviceid(its_cmd);
+       bool valid = its_cmd_get_validbit(its_cmd);
+       struct its_device *device;
+
+       if (!vgic_its_check_device_id(kvm, its, device_id))
+               return E_ITS_MAPD_DEVICE_OOR;
+
+       device = find_its_device(its, device_id);
+
+       /*
+        * The spec says that calling MAPD on an already mapped device
+        * invalidates all cached data for this device. We implement this
+        * by removing the mapping and re-establishing it.
+        */
+       if (device)
+               vgic_its_unmap_device(kvm, device);
+
+       /*
+        * The spec does not say whether unmapping a not-mapped device
+        * is an error, so we are done in any case.
+        */
+       if (!valid)
+               return 0;
+
+       device = kzalloc(sizeof(struct its_device), GFP_KERNEL);
+       if (!device)
+               return -ENOMEM;
+
+       device->device_id = device_id;
+       INIT_LIST_HEAD(&device->itt_head);
+
+       list_add_tail(&device->dev_list, &its->device_list);
+
+       return 0;
+}
+
+static int vgic_its_nr_collection_ids(struct vgic_its *its)
+{
+       u64 r = its->baser_coll_table;
+
+       return (GITS_BASER_NR_PAGES(r) * SZ_64K) / GITS_BASER_ENTRY_SIZE(r);
+}
+
+/*
+ * The MAPC command maps collection IDs to redistributors.
+ * Must be called with the its_lock mutex held.
+ */
+static int vgic_its_cmd_handle_mapc(struct kvm *kvm, struct vgic_its *its,
+                                   u64 *its_cmd)
+{
+       u16 coll_id;
+       u32 target_addr;
+       struct its_collection *collection;
+       bool valid;
+
+       valid = its_cmd_get_validbit(its_cmd);
+       coll_id = its_cmd_get_collection(its_cmd);
+       target_addr = its_cmd_get_target_addr(its_cmd);
+
+       if (target_addr >= atomic_read(&kvm->online_vcpus))
+               return E_ITS_MAPC_PROCNUM_OOR;
+
+       if (coll_id >= vgic_its_nr_collection_ids(its))
+               return E_ITS_MAPC_COLLECTION_OOR;
+
+       collection = find_collection(its, coll_id);
+
+       if (!valid) {
+               struct its_device *device;
+               struct its_itte *itte;
+               /*
+                * Clearing the mapping for that collection ID removes the
+                * entry from the list. If there wasn't any before, we can
+                * go home early.
+                */
+               if (!collection)
+                       return 0;
+
+               for_each_lpi_its(device, itte, its)
+                       if (itte->collection &&
+                           itte->collection->collection_id == coll_id)
+                               itte->collection = NULL;
+
+               list_del(&collection->coll_list);
+               kfree(collection);
+       } else {
+               if (!collection) {
+                       collection = kzalloc(sizeof(struct its_collection),
+                                            GFP_KERNEL);
+                       if (!collection)
+                               return -ENOMEM;
+
+                       vgic_its_init_collection(its, collection, coll_id);
+                       collection->target_addr = target_addr;
+               } else {
+                       collection->target_addr = target_addr;
+                       update_affinity_collection(kvm, its, collection);
+               }
+       }
+
+       return 0;
+}
+
+/*
+ * The CLEAR command removes the pending state for a particular LPI.
+ * Must be called with the its_lock mutex held.
+ */
+static int vgic_its_cmd_handle_clear(struct kvm *kvm, struct vgic_its *its,
+                                    u64 *its_cmd)
+{
+       u32 device_id = its_cmd_get_deviceid(its_cmd);
+       u32 event_id = its_cmd_get_id(its_cmd);
+       struct its_itte *itte;
+
+
+       itte = find_itte(its, device_id, event_id);
+       if (!itte)
+               return E_ITS_CLEAR_UNMAPPED_INTERRUPT;
+
+       itte->irq->pending = false;
+
+       return 0;
+}
+
+/*
+ * The INV command syncs the configuration bits from the memory table.
+ * Must be called with the its_lock mutex held.
+ */
+static int vgic_its_cmd_handle_inv(struct kvm *kvm, struct vgic_its *its,
+                                  u64 *its_cmd)
+{
+       u32 device_id = its_cmd_get_deviceid(its_cmd);
+       u32 event_id = its_cmd_get_id(its_cmd);
+       struct its_itte *itte;
+
+
+       itte = find_itte(its, device_id, event_id);
+       if (!itte)
+               return E_ITS_INV_UNMAPPED_INTERRUPT;
+
+       return update_lpi_config(kvm, itte->irq, NULL);
+}
+
+/*
+ * The INVALL command requests flushing of all IRQ data in this collection.
+ * Find the VCPU mapped to that collection, then iterate over the VM's list
+ * of mapped LPIs and update the configuration for each IRQ which targets
+ * the specified vcpu. The configuration will be read from the in-memory
+ * configuration table.
+ * Must be called with the its_lock mutex held.
+ */
+static int vgic_its_cmd_handle_invall(struct kvm *kvm, struct vgic_its *its,
+                                     u64 *its_cmd)
+{
+       u32 coll_id = its_cmd_get_collection(its_cmd);
+       struct its_collection *collection;
+       struct kvm_vcpu *vcpu;
+       struct vgic_irq *irq;
+       u32 *intids;
+       int irq_count, i;
+
+       collection = find_collection(its, coll_id);
+       if (!its_is_collection_mapped(collection))
+               return E_ITS_INVALL_UNMAPPED_COLLECTION;
+
+       vcpu = kvm_get_vcpu(kvm, collection->target_addr);
+
+       irq_count = vgic_copy_lpi_list(kvm, &intids);
+       if (irq_count < 0)
+               return irq_count;
+
+       for (i = 0; i < irq_count; i++) {
+               irq = vgic_get_irq(kvm, NULL, intids[i]);
+               if (!irq)
+                       continue;
+               update_lpi_config(kvm, irq, vcpu);
+               vgic_put_irq(kvm, irq);
+       }
+
+       kfree(intids);
+
+       return 0;
+}
+
+/*
+ * The MOVALL command moves the pending state of all IRQs targeting one
+ * redistributor to another. We don't hold the pending state in the VCPUs,
+ * but in the IRQs instead, so there is really not much to do for us here.
+ * However the spec says that no IRQ must target the old redistributor
+ * afterwards, so we make sure that no LPI is using the associated target_vcpu.
+ * This command affects all LPIs in the system that target that redistributor.
+ */
+static int vgic_its_cmd_handle_movall(struct kvm *kvm, struct vgic_its *its,
+                                     u64 *its_cmd)
+{
+       struct vgic_dist *dist = &kvm->arch.vgic;
+       u32 target1_addr = its_cmd_get_target_addr(its_cmd);
+       u32 target2_addr = its_cmd_mask_field(its_cmd, 3, 16, 32);
+       struct kvm_vcpu *vcpu1, *vcpu2;
+       struct vgic_irq *irq;
+
+       if (target1_addr >= atomic_read(&kvm->online_vcpus) ||
+           target2_addr >= atomic_read(&kvm->online_vcpus))
+               return E_ITS_MOVALL_PROCNUM_OOR;
+
+       if (target1_addr == target2_addr)
+               return 0;
+
+       vcpu1 = kvm_get_vcpu(kvm, target1_addr);
+       vcpu2 = kvm_get_vcpu(kvm, target2_addr);
+
+       spin_lock(&dist->lpi_list_lock);
+
+       list_for_each_entry(irq, &dist->lpi_list_head, lpi_list) {
+               spin_lock(&irq->irq_lock);
+
+               if (irq->target_vcpu == vcpu1)
+                       irq->target_vcpu = vcpu2;
+
+               spin_unlock(&irq->irq_lock);
+       }
+
+       spin_unlock(&dist->lpi_list_lock);
+
+       return 0;
+}
+
+/*
+ * This function is called with the its_cmd lock held, but the ITS data
+ * structure lock dropped.
+ */
 static int vgic_its_handle_command(struct kvm *kvm, struct vgic_its *its,
                                   u64 *its_cmd)
 {
-       return -ENODEV;
+       u8 cmd = its_cmd_get_command(its_cmd);
+       int ret = -ENODEV;
+
+       mutex_lock(&its->its_lock);
+       switch (cmd) {
+       case GITS_CMD_MAPD:
+               ret = vgic_its_cmd_handle_mapd(kvm, its, its_cmd);
+               break;
+       case GITS_CMD_MAPC:
+               ret = vgic_its_cmd_handle_mapc(kvm, its, its_cmd);
+               break;
+       case GITS_CMD_MAPI:
+               ret = vgic_its_cmd_handle_mapi(kvm, its, its_cmd, cmd);
+               break;
+       case GITS_CMD_MAPTI:
+               ret = vgic_its_cmd_handle_mapi(kvm, its, its_cmd, cmd);
+               break;
+       case GITS_CMD_MOVI:
+               ret = vgic_its_cmd_handle_movi(kvm, its, its_cmd);
+               break;
+       case GITS_CMD_DISCARD:
+               ret = vgic_its_cmd_handle_discard(kvm, its, its_cmd);
+               break;
+       case GITS_CMD_CLEAR:
+               ret = vgic_its_cmd_handle_clear(kvm, its, its_cmd);
+               break;
+       case GITS_CMD_MOVALL:
+               ret = vgic_its_cmd_handle_movall(kvm, its, its_cmd);
+               break;
+       case GITS_CMD_INV:
+               ret = vgic_its_cmd_handle_inv(kvm, its, its_cmd);
+               break;
+       case GITS_CMD_INVALL:
+               ret = vgic_its_cmd_handle_invall(kvm, its, its_cmd);
+               break;
+       case GITS_CMD_SYNC:
+               /* we ignore this command: we are in sync all of the time */
+               ret = 0;
+               break;
+       }
+       mutex_unlock(&its->its_lock);
+
+       return ret;
 }
 
 static u64 vgic_sanitise_its_baser(u64 reg)