diff --git a/src/gallium/drivers/zink/zink_kopper.c b/src/gallium/drivers/zink/zink_kopper.c
index 1a19e62497c..99a5c8005e0 100644
--- a/src/gallium/drivers/zink/zink_kopper.c
+++ b/src/gallium/drivers/zink/zink_kopper.c
@@ -954,8 +954,21 @@ zink_kopper_acquire_readback(struct zink_context *ctx, struct zink_resource *res
    }
    while (res->obj->dt_idx != last_dt_idx) {
       cdt->age_locked = true;
-      if (res->obj->dt_idx != UINT32_MAX && !zink_kopper_present_readback(ctx, res))
-         break;
+      if (res->obj->dt_idx != UINT32_MAX) {
+         if (!zink_kopper_present_readback(ctx, res))
+            break;
+      } else if (util_queue_is_initialized(&screen->flush_queue)) {
+         /* AcquireNextImageKHR and QueuePresentKHR both access the swapchain, and
+          * if res->obj->dt_idx == UINT32_MAX then zink_kopper_present_readback is
+          * not called and we don't wait for the cdt->swapchain->present_fence.
+          * Still, a kopper_present might have been called in another thread, like
+          * e.g. with spec@!opengl 1.1@read-front, so we have to wait until the
+          * last call to QueuePresentKHR is finished to avoid an
+          *    UNASSIGNED-Threading-MultipleThreads-Write
+          * validation error that indicats a race condition when accessing the swapchain.
+          */
+         util_queue_fence_wait(&cdt->swapchain->present_fence);
+      }
       cdt->age_locked = true;
       do {
          ret = kopper_acquire(screen, res, 0);