diff --git a/src/freedreno/vulkan/tu_knl_drm_msm.cc b/src/freedreno/vulkan/tu_knl_drm_msm.cc
index 3bf4ada3457..e3a69161105 100644
--- a/src/freedreno/vulkan/tu_knl_drm_msm.cc
+++ b/src/freedreno/vulkan/tu_knl_drm_msm.cc
@@ -335,30 +335,85 @@ msm_device_check_status(struct tu_device *device)
 static int
 msm_submitqueue_new(struct tu_device *dev, struct tu_queue *queue)
 {
-   assert(queue->priority >= 0 &&
-          queue->priority < dev->physical_device->submitqueue_priority_count);
-   struct drm_msm_submitqueue req = {
-      .flags = queue->type == TU_QUEUE_SPARSE ? MSM_SUBMITQUEUE_VM_BIND :
-            (dev->physical_device->info->chip >= 7 &&
-             dev->physical_device->has_preemption ?
-             MSM_SUBMITQUEUE_ALLOW_PREEMPT : 0),
-      .prio = queue->priority,
-   };
+   int ret = 0;
 
-   int ret = drmCommandWriteRead(dev->fd,
-                                 DRM_MSM_SUBMITQUEUE_NEW, &req, sizeof(req));
-   if (ret)
-      return ret;
+   if (queue->type == TU_QUEUE_GFX) {
+      assert(queue->priority >= 0 &&
+             queue->priority < dev->physical_device->submitqueue_priority_count);
+      struct drm_msm_submitqueue req = {
+         .flags = (dev->physical_device->info->chip >= 7 &&
+                   dev->physical_device->has_preemption ?
+                   MSM_SUBMITQUEUE_ALLOW_PREEMPT : 0),
+         .prio = queue->priority,
+      };
+
+      ret = drmCommandWriteRead(dev->fd,
+                                DRM_MSM_SUBMITQUEUE_NEW, &req, sizeof(req));
+      if (ret)
+         return ret;
+
+      queue->msm_queue_id = req.id;
+   }
+
+   if (queue->type == TU_QUEUE_SPARSE ||
+       dev->physical_device->has_vm_bind) {
+      struct drm_msm_submitqueue req = {
+         .flags = MSM_SUBMITQUEUE_VM_BIND,
+         .prio = 0,
+      };
+
+      ret = drmCommandWriteRead(dev->fd,
+                                DRM_MSM_SUBMITQUEUE_NEW, &req, sizeof(req));
+      if (ret)
+         goto fail_sparse;
+
+      queue->sparse_queue_id = req.id;
+   }
+
+   if (queue->type == TU_QUEUE_GFX && dev->physical_device->has_vm_bind) {
+      ret = drmSyncobjCreate(dev->fd, 0, &queue->gfx_syncobj);
+      if (ret)
+         goto fail_gfx_syncobj;
+      ret = drmSyncobjCreate(dev->fd, 0, &queue->sparse_syncobj);
+      if (ret)
+         goto fail_sparse_syncobj;
+   }
 
-   queue->msm_queue_id = req.id;
    return 0;
+
+fail_sparse_syncobj:
+   drmSyncobjDestroy(dev->fd, queue->gfx_syncobj);
+fail_gfx_syncobj:
+   if (queue->type == TU_QUEUE_SPARSE || dev->physical_device->has_vm_bind) {
+      drmCommandWrite(dev->fd, DRM_MSM_SUBMITQUEUE_CLOSE,
+                      &queue->sparse_queue_id, sizeof(uint32_t));
+   }
+fail_sparse:
+   if (queue->type == TU_QUEUE_GFX) {
+      drmCommandWrite(dev->fd, DRM_MSM_SUBMITQUEUE_CLOSE,
+                      &queue->msm_queue_id, sizeof(uint32_t));
+   }
+
+   return ret;
 }
 
 static void
 msm_submitqueue_close(struct tu_device *dev, struct tu_queue *queue)
 {
-   drmCommandWrite(dev->fd, DRM_MSM_SUBMITQUEUE_CLOSE,
-                   &queue->msm_queue_id, sizeof(uint32_t));
+   if (queue->msm_queue_id) {
+      drmCommandWrite(dev->fd, DRM_MSM_SUBMITQUEUE_CLOSE,
+                      &queue->msm_queue_id, sizeof(uint32_t));
+   }
+
+   if (queue->sparse_queue_id) {
+      drmCommandWrite(dev->fd, DRM_MSM_SUBMITQUEUE_CLOSE,
+                      &queue->sparse_queue_id, sizeof(uint32_t));
+   }
+
+   if (queue->type == TU_QUEUE_GFX && dev->physical_device->has_vm_bind) {
+      drmSyncobjDestroy(dev->fd, queue->sparse_syncobj);
+      drmSyncobjDestroy(dev->fd, queue->gfx_syncobj);
+   }
 }
 
 static void
@@ -1199,11 +1254,18 @@ msm_queue_submit(struct tu_queue *queue, void *_submit,
    uint64_t start_ts = tu_perfetto_begin_submit();
 #endif
    uint32_t fence = 0;
+   uint64_t other_timepoint =
+      submit->binds.size != 0 ? queue->gfx_timepoint :
+      queue->sparse_timepoint;
+   uint32_t extra_wait_count =
+      (queue->type == TU_QUEUE_GFX && other_timepoint != 0) ? 1 : 0;
+   uint32_t extra_signal_count =
+      (queue->type == TU_QUEUE_GFX && has_vm_bind) ? 1 : 0;
 
    /* Allocate without wait timeline semaphores */
    in_syncobjs = (struct drm_msm_syncobj *) vk_zalloc(
       &queue->device->vk.alloc,
-      wait_count * sizeof(*in_syncobjs), 8,
+      (wait_count + extra_wait_count) * sizeof(*in_syncobjs), 8,
       VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
 
    if (in_syncobjs == NULL) {
@@ -1214,7 +1276,7 @@ msm_queue_submit(struct tu_queue *queue, void *_submit,
    /* Allocate with signal timeline semaphores considered */
    out_syncobjs = (struct drm_msm_syncobj *) vk_zalloc(
       &queue->device->vk.alloc,
-      signal_count * sizeof(*out_syncobjs), 8,
+      (signal_count + extra_signal_count) * sizeof(*out_syncobjs), 8,
       VK_SYSTEM_ALLOCATION_SCOPE_DEVICE);
 
    if (out_syncobjs == NULL) {
@@ -1242,7 +1304,22 @@ msm_queue_submit(struct tu_queue *queue, void *_submit,
       };
    }
 
-   if (queue->type == TU_QUEUE_SPARSE) {
+   if (queue->type == TU_QUEUE_SPARSE || submit->binds.size != 0) {
+      if (extra_wait_count) {
+         in_syncobjs[wait_count] = (struct drm_msm_syncobj) {
+            .handle = queue->gfx_syncobj,
+            .flags = 0,
+            .point = queue->gfx_timepoint,
+         };
+      }
+      if (extra_signal_count) {
+         out_syncobjs[signal_count] = (struct drm_msm_syncobj) {
+            .handle = queue->sparse_syncobj,
+            .flags = 0,
+            .point = ++queue->sparse_timepoint,
+         };
+      }
+
       unsigned nr_ops = util_dynarray_num_elements(&submit->binds,
                                                    struct drm_msm_vm_bind_op);
 
@@ -1271,11 +1348,11 @@ msm_queue_submit(struct tu_queue *queue, void *_submit,
          .flags = flags,
          .nr_ops = nr_ops,
          .fence_fd = queue->device->vm_bind_fence_fd,
-         .queue_id = queue->msm_queue_id,
+         .queue_id = queue->sparse_queue_id,
          .in_syncobjs = (uint64_t)(uintptr_t)in_syncobjs,
          .out_syncobjs = (uint64_t)(uintptr_t)out_syncobjs,
-         .nr_in_syncobjs = wait_count,
-         .nr_out_syncobjs = signal_count,
+         .nr_in_syncobjs = wait_count + extra_wait_count,
+         .nr_out_syncobjs = signal_count + extra_signal_count,
          .syncobj_stride = sizeof(struct drm_msm_syncobj),
          .op_stride = sizeof(struct drm_msm_vm_bind_op),
       };
@@ -1337,9 +1414,13 @@ msm_queue_submit(struct tu_queue *queue, void *_submit,
                   .flags = flags,
                   .nr_ops = 1,
                   .fence_fd = queue->device->vm_bind_fence_fd,
-                  .queue_id = queue->msm_queue_id,
+                  .queue_id = queue->sparse_queue_id,
+                  .in_syncobjs =
+                     (uint64_t)(uintptr_t)&in_syncobjs[wait_count],
                   .out_syncobjs = (uint64_t)(uintptr_t)out_syncobjs,
-                  .nr_out_syncobjs = last ? signal_count : 0,
+                  .nr_in_syncobjs = extra_wait_count,
+                  .nr_out_syncobjs = last ? signal_count +
+                     extra_signal_count : 0,
                   .syncobj_stride = sizeof(struct drm_msm_syncobj),
                   .op_stride = sizeof(struct drm_msm_vm_bind_op),
                   .op = *op,
@@ -1369,6 +1450,21 @@ msm_queue_submit(struct tu_queue *queue, void *_submit,
          flags |= MSM_SUBMIT_SYNCOBJ_OUT;
 
       if (has_vm_bind) {
+         if (extra_wait_count) {
+            in_syncobjs[wait_count] = (struct drm_msm_syncobj) {
+               .handle = queue->sparse_syncobj,
+               .flags = 0,
+               .point = queue->sparse_timepoint,
+            };
+         }
+         if (extra_signal_count) {
+            out_syncobjs[signal_count] = (struct drm_msm_syncobj) {
+               .handle = queue->gfx_syncobj,
+               .flags = 0,
+               .point = ++queue->gfx_timepoint,
+            };
+         }
+
          u_rwlock_rdlock(&queue->device->vm_bind_fence_lock);
 
          if (queue->device->vm_bind_fence_fd != -1)
@@ -1422,8 +1518,8 @@ msm_queue_submit(struct tu_queue *queue, void *_submit,
          .queueid = queue->msm_queue_id,
          .in_syncobjs = (uint64_t)(uintptr_t)in_syncobjs,
          .out_syncobjs = (uint64_t)(uintptr_t)out_syncobjs,
-         .nr_in_syncobjs = wait_count,
-         .nr_out_syncobjs = signal_count,
+         .nr_in_syncobjs = wait_count + extra_wait_count,
+         .nr_out_syncobjs = signal_count + extra_signal_count,
          .syncobj_stride = sizeof(struct drm_msm_syncobj),
       };
 
diff --git a/src/freedreno/vulkan/tu_queue.h b/src/freedreno/vulkan/tu_queue.h
index b5fccabcc2e..3f7f78a92ba 100644
--- a/src/freedreno/vulkan/tu_queue.h
+++ b/src/freedreno/vulkan/tu_queue.h
@@ -28,6 +28,11 @@ struct tu_queue
    uint32_t msm_queue_id;
    uint32_t priority;
 
+   uint32_t sparse_queue_id;
+
+   uint32_t sparse_syncobj, gfx_syncobj;
+   uint64_t sparse_timepoint, gfx_timepoint;
+
    int fence;           /* timestamp/fence of the last queue submission */
 };
 VK_DEFINE_HANDLE_CASTS(tu_queue, vk.base, VkQueue, VK_OBJECT_TYPE_QUEUE)