diff --git a/src/freedreno/vulkan/tu_knl_drm_msm.cc b/src/freedreno/vulkan/tu_knl_drm_msm.cc
index 5388b321e2b..749e0593651 100644
--- a/src/freedreno/vulkan/tu_knl_drm_msm.cc
+++ b/src/freedreno/vulkan/tu_knl_drm_msm.cc
@@ -805,6 +805,7 @@ msm_queue_submit(struct tu_queue *queue, void *_submit,
       util_dynarray_num_elements(&submit->commands, struct drm_msm_gem_submit_cmd);
 #if HAVE_PERFETTO
    struct tu_perfetto_clocks clocks;
+   uint64_t start_ts = tu_perfetto_begin_submit();
 #endif
 
    uint32_t flags = MSM_PIPE_3D0;
@@ -904,7 +905,8 @@ msm_queue_submit(struct tu_queue *queue, void *_submit,
    p_atomic_set(&queue->fence, req.fence);
 
 #if HAVE_PERFETTO
-   clocks = tu_perfetto_submit(queue->device, queue->device->submit_count, NULL);
+   clocks = tu_perfetto_end_submit(queue, queue->device->submit_count,
+                                   start_ts, NULL);
    gpu_offset = clocks.gpu_ts_offset;
 #endif
 
diff --git a/src/freedreno/vulkan/tu_knl_drm_virtio.cc b/src/freedreno/vulkan/tu_knl_drm_virtio.cc
index c4630f3ea48..9895b704634 100644
--- a/src/freedreno/vulkan/tu_knl_drm_virtio.cc
+++ b/src/freedreno/vulkan/tu_knl_drm_virtio.cc
@@ -881,6 +881,7 @@ virtio_queue_submit(struct tu_queue *queue, void *_submit,
    struct vdrm_execbuf_params params;
 #if HAVE_PERFETTO
    struct tu_perfetto_clocks clocks;
+   uint64_t start_ts = tu_perfetto_begin_submit();
 #endif
 
    /* It would be nice to not need to defer this, but virtio_device_init()
@@ -1024,7 +1025,8 @@ virtio_queue_submit(struct tu_queue *queue, void *_submit,
    }
 
 #if HAVE_PERFETTO
-   clocks = tu_perfetto_submit(queue->device, queue->device->submit_count, NULL);
+   clocks = tu_perfetto_end_submit(queue, queue->device->submit_count,
+                                   start_ts, NULL);
    gpu_offset = clocks.gpu_ts_offset;
 #endif
 
diff --git a/src/freedreno/vulkan/tu_knl_kgsl.cc b/src/freedreno/vulkan/tu_knl_kgsl.cc
index b6e17fd4ebd..74288a6c7a5 100644
--- a/src/freedreno/vulkan/tu_knl_kgsl.cc
+++ b/src/freedreno/vulkan/tu_knl_kgsl.cc
@@ -1084,6 +1084,10 @@ kgsl_queue_submit(struct tu_queue *queue, void *_submit,
    struct tu_kgsl_queue_submit *submit =
       (struct tu_kgsl_queue_submit *)_submit;
 
+#if HAVE_PERFETTO
+   uint64_t start_ts = tu_perfetto_begin_submit();
+#endif
+
    if (submit->commands.size == 0) {
       const struct kgsl_syncobj *wait_semaphores[wait_count + 1];
       for (uint32_t i = 0; i < wait_count; i++) {
@@ -1257,7 +1261,8 @@ kgsl_queue_submit(struct tu_queue *queue, void *_submit,
          .gpu_ts_offset = gpu_offset,
       };
 
-      clocks = tu_perfetto_submit(queue->device, queue->device->submit_count, &clocks);
+      clocks = tu_perfetto_end_submit(queue, queue->device->submit_count,
+                                      start_ts, &clocks);
       gpu_offset = clocks.gpu_ts_offset;
    }
 #endif
diff --git a/src/freedreno/vulkan/tu_perfetto.cc b/src/freedreno/vulkan/tu_perfetto.cc
index 93afa1f9bd1..a428159d51e 100644
--- a/src/freedreno/vulkan/tu_perfetto.cc
+++ b/src/freedreno/vulkan/tu_perfetto.cc
@@ -8,6 +8,7 @@
 #include "tu_perfetto.h"
 #include "tu_buffer.h"
 #include "tu_device.h"
+#include "tu_queue.h"
 #include "tu_image.h"
 
 #include "util/hash_table.h"
@@ -363,34 +364,21 @@ emit_sync_timestamp(uint64_t cpu_ts, uint64_t gpu_ts)
    });
 }
 
-static void
-emit_submit_id(uint32_t submission_id)
+uint64_t
+tu_perfetto_begin_submit()
 {
-   TuRenderpassDataSource::Trace([=](TuRenderpassDataSource::TraceContext tctx) {
-      auto packet = tctx.NewTracePacket();
-
-      packet->set_timestamp(perfetto::base::GetBootTimeNs().count());
-
-      auto event = packet->set_vulkan_api_event();
-      auto submit = event->set_vk_queue_submit();
-
-      submit->set_submission_id(submission_id);
-   });
+   return perfetto::base::GetBootTimeNs().count();
 }
 
-struct tu_perfetto_clocks
-tu_perfetto_submit(struct tu_device *dev,
-                   uint32_t submission_id,
-                   struct tu_perfetto_clocks *gpu_clocks)
+static struct tu_perfetto_clocks
+sync_clocks(struct tu_device *dev,
+            const struct tu_perfetto_clocks *gpu_clocks)
 {
    struct tu_perfetto_clocks clocks {};
    if (gpu_clocks) {
       clocks = *gpu_clocks;
    }
 
-   if (!u_trace_perfetto_active(tu_device_get_u_trace(dev)))
-      return {};
-
    clocks.cpu = perfetto::base::GetBootTimeNs().count();
 
    if (gpu_clocks) {
@@ -456,8 +444,36 @@ tu_perfetto_submit(struct tu_device *dev,
       next_clock_sync_ns = clocks.cpu + 30000000;
    }
 
-   emit_sync_timestamp(clocks.cpu, clocks.gpu_ts + clocks.gpu_ts_offset);
-   emit_submit_id(submission_id);
+   return clocks;
+}
+
+struct tu_perfetto_clocks
+tu_perfetto_end_submit(struct tu_queue *queue,
+                       uint32_t submission_id,
+                       uint64_t start_ts,
+                       struct tu_perfetto_clocks *gpu_clocks)
+{
+   struct tu_device *dev = queue->device;
+   if (!u_trace_perfetto_active(tu_device_get_u_trace(dev)))
+      return {};
+
+   struct tu_perfetto_clocks clocks = sync_clocks(dev, gpu_clocks);
+   if (clocks.gpu_ts > 0)
+      emit_sync_timestamp(clocks.cpu, clocks.gpu_ts + clocks.gpu_ts_offset);
+
+   TuRenderpassDataSource::Trace([=](TuRenderpassDataSource::TraceContext tctx) {
+      auto packet = tctx.NewTracePacket();
+
+      packet->set_timestamp(start_ts);
+
+      auto event = packet->set_vulkan_api_event();
+      auto submit = event->set_vk_queue_submit();
+
+      submit->set_duration_ns(clocks.cpu - start_ts);
+      submit->set_vk_queue((uintptr_t) queue);
+      submit->set_submission_id(submission_id);
+   });
+
    return clocks;
 }
 
diff --git a/src/freedreno/vulkan/tu_perfetto.h b/src/freedreno/vulkan/tu_perfetto.h
index 6ed86fe9761..6068d6e190a 100644
--- a/src/freedreno/vulkan/tu_perfetto.h
+++ b/src/freedreno/vulkan/tu_perfetto.h
@@ -18,6 +18,7 @@ extern "C" {
 #define TU_PERFETTO_MAX_STACK_DEPTH 8
 
 struct tu_device;
+struct tu_queue;
 struct tu_u_trace_submission_data;
 
 struct tu_perfetto_stage {
@@ -46,10 +47,14 @@ struct tu_perfetto_clocks
    uint64_t gpu_ts_offset;
 };
 
+uint64_t
+tu_perfetto_begin_submit();
+
 struct tu_perfetto_clocks
-tu_perfetto_submit(struct tu_device *dev,
-                   uint32_t submission_id,
-                   struct tu_perfetto_clocks *clocks);
+tu_perfetto_end_submit(struct tu_queue *queue,
+                       uint32_t submission_id,
+                       uint64_t start_ts,
+                       struct tu_perfetto_clocks *clocks);
 
 void tu_perfetto_log_create_buffer(struct tu_device *dev, struct tu_buffer *buffer);
 void tu_perfetto_log_bind_buffer(struct tu_device *dev, struct tu_buffer *buffer);