[PATCH v4 14/14] vfio/migration: Optimize vfio_save_pending()

[Avihai Horon]

During pre-copy phase of migration vfio_save_pending() is called
repeatedly and queries the VFIO device for its pending data size.

As long as pending RAM size is over the threshold, migration can't
converge and be completed. Therefore, during this time there is no
point in querying the VFIO device pending data size.

Avoid these unnecessary queries by issuing them in a RAM pre-copy
notifier instead of vfio_save_pending().

This way the VFIO device is queried only when RAM pending data is
below the threshold, when there is an actual chance for migration to
converge.

Signed-off-by: Avihai Horon <avihaih@nvidia.com>
---
 hw/vfio/migration.c           | 55 +++++++++++++++++++++++++----------
 hw/vfio/trace-events          |  1 +
 include/hw/vfio/vfio-common.h |  2 ++
 3 files changed, 43 insertions(+), 15 deletions(-)

diff --git a/hw/vfio/migration.c b/hw/vfio/migration.c
index 9285746183..d57cda5516 100644
--- a/hw/vfio/migration.c
+++ b/hw/vfio/migration.c
@@ -269,31 +269,19 @@ static void vfio_save_cleanup(void *opaque)
     trace_vfio_save_cleanup(vbasedev->name);
 }
 
-/*
- * Migration size of VFIO devices can be as little as a few KBs or as big as
- * many GBs. This value should be big enough to cover the worst case.
- */
-#define VFIO_MIG_STOP_COPY_SIZE (100 * GiB)
 static void vfio_save_pending(void *opaque, uint64_t threshold_size,
                               uint64_t *res_precopy_only,
                               uint64_t *res_compatible,
                               uint64_t *res_postcopy_only)
 {
     VFIODevice *vbasedev = opaque;
-    uint64_t stop_copy_size;
+    VFIOMigration *migration = vbasedev->migration;
 
-    if (vfio_query_stop_copy_size(vbasedev, &stop_copy_size)) {
-        /*
-         * Failed to get pending migration size. Report big pending size so
-         * downtime limit won't be violated.
-         */
-        stop_copy_size = VFIO_MIG_STOP_COPY_SIZE;
-    }
+    *res_precopy_only += migration->stop_copy_size;
 
-    *res_precopy_only += stop_copy_size;
     trace_vfio_save_pending(vbasedev->name, *res_precopy_only,
                             *res_postcopy_only, *res_compatible,
-                            stop_copy_size);
+                            migration->stop_copy_size);
 }
 
 /* Returns 1 if end-of-stream is reached, 0 if more data and -1 if error */
@@ -501,6 +489,40 @@ static void vfio_migration_state_notifier(Notifier 
*notifier, void *data)
     }
 }
 
+/*
+ * Migration size of VFIO devices can be as little as a few KBs or as big as
+ * many GBs. This value should be big enough to cover the worst case.
+ */
+#define VFIO_MIG_STOP_COPY_SIZE (100 * GiB)
+static int vfio_migration_data_notifier(NotifierWithReturn *n, void *data)
+{
+    VFIOMigration *migration = container_of(n, VFIOMigration, migration_data);
+    VFIODevice *vbasedev = migration->vbasedev;
+    PrecopyNotifyData *pnd = data;
+
+    if (pnd->reason != PRECOPY_NOTIFY_AFTER_BITMAP_SYNC) {
+        return 0;
+    }
+
+    /* No need to get pending size when finishing migration */
+    if (runstate_check(RUN_STATE_FINISH_MIGRATE)) {
+        return 0;
+    }
+
+    if (vfio_query_stop_copy_size(vbasedev, &migration->stop_copy_size)) {
+        /*
+         * Failed to get pending migration size. Report big pending size so
+         * downtime limit won't be violated.
+         */
+        migration->stop_copy_size = VFIO_MIG_STOP_COPY_SIZE;
+    }
+
+    trace_vfio_migration_data_notifier(vbasedev->name,
+                                       migration->stop_copy_size);
+
+    return 0;
+}
+
 static void vfio_migration_exit(VFIODevice *vbasedev)
 {
     g_free(vbasedev->migration);
@@ -578,6 +600,8 @@ static int vfio_migration_init(VFIODevice *vbasedev)
                                                            vbasedev);
     migration->migration_state.notify = vfio_migration_state_notifier;
     add_migration_state_change_notifier(&migration->migration_state);
+    migration->migration_data.notify = vfio_migration_data_notifier;
+    precopy_add_notifier(&migration->migration_data);
 
     return 0;
 }
@@ -622,6 +646,7 @@ void vfio_migration_finalize(VFIODevice *vbasedev)
     if (vbasedev->migration) {
         VFIOMigration *migration = vbasedev->migration;
 
+        precopy_remove_notifier(&migration->migration_data);
         remove_migration_state_change_notifier(&migration->migration_state);
         qemu_del_vm_change_state_handler(migration->vm_state);
         unregister_savevm(VMSTATE_IF(vbasedev->dev), "vfio", vbasedev);
diff --git a/hw/vfio/trace-events b/hw/vfio/trace-events
index 2723a5d1aa..e377a24f0e 100644
--- a/hw/vfio/trace-events
+++ b/hw/vfio/trace-events
@@ -154,6 +154,7 @@ vfio_load_cleanup(const char *name) " (%s)"
 vfio_load_device_config_state(const char *name) " (%s)"
 vfio_load_state(const char *name, uint64_t data) " (%s) data 0x%"PRIx64
 vfio_load_state_device_data(const char *name, uint64_t data_size, int ret) " 
(%s) size 0x%"PRIx64" ret %d"
+vfio_migration_data_notifier(const char *name, uint64_t stopcopy_size) " (%s) 
stopcopy size 0x%"PRIx64
 vfio_migration_probe(const char *name) " (%s)"
 vfio_migration_set_state(const char *name, const char *state) " (%s) state %s"
 vfio_migration_state_notifier(const char *name, const char *state) " (%s) 
state %s"
diff --git a/include/hw/vfio/vfio-common.h b/include/hw/vfio/vfio-common.h
index 76d470178f..2aba45887c 100644
--- a/include/hw/vfio/vfio-common.h
+++ b/include/hw/vfio/vfio-common.h
@@ -62,10 +62,12 @@ typedef struct VFIOMigration {
     struct VFIODevice *vbasedev;
     VMChangeStateEntry *vm_state;
     Notifier migration_state;
+    NotifierWithReturn migration_data;
     enum vfio_device_mig_state device_state;
     int data_fd;
     void *data_buffer;
     size_t data_buffer_size;
+    uint64_t stop_copy_size;
 } VFIOMigration;
 
 typedef struct VFIOAddressSpace {
-- 
2.26.3