diff --git a/src/intel/vulkan/anv_batch_chain.c b/src/intel/vulkan/anv_batch_chain.c
index 15677320c4d..a32eb3bd9ec 100644
--- a/src/intel/vulkan/anv_batch_chain.c
+++ b/src/intel/vulkan/anv_batch_chain.c
@@ -2264,6 +2264,12 @@ anv_queue_exec_locked(struct anv_queue *queue,
    return result;
 }
 
+static inline bool
+can_chain_query_pools(struct anv_query_pool *p1, struct anv_query_pool *p2)
+{
+   return (!p1 || !p2 || p1 == p2);
+}
+
 static VkResult
 anv_queue_submit_locked(struct anv_queue *queue,
                         struct vk_queue_submit *submit)
@@ -2280,55 +2286,49 @@ anv_queue_submit_locked(struct anv_queue *queue,
       if (result != VK_SUCCESS)
          return result;
    } else {
-      struct anv_query_pool *perf_query_pool = NULL;
-      uint32_t start = 0;
-
       /* Everything's easier if we don't have to bother with container_of() */
       STATIC_ASSERT(offsetof(struct anv_cmd_buffer, vk) == 0);
       struct vk_command_buffer **vk_cmd_buffers = submit->command_buffers;
       struct anv_cmd_buffer **cmd_buffers = (void *)vk_cmd_buffers;
-
-      const uint32_t end = submit->command_buffer_count;
-      while (start < end) {
-         uint32_t i = start + 1;
-
-         /* Save the first query pool or NULL */
-         perf_query_pool = cmd_buffers[start]->perf_query_pool;
-
-         for (; i < end; i++) {
-            /* Can we chain the last buffer into the next one? */
-            if (!anv_cmd_buffer_is_chainable(cmd_buffers[i]))
-               break;
-
-            if (cmd_buffers[i]->perf_query_pool != NULL) {
-               if (perf_query_pool != NULL) {
-                  /* They have to have the same query pool */
-                  if (cmd_buffers[i]->perf_query_pool != perf_query_pool)
-                     break;
-               } else {
-                  perf_query_pool = cmd_buffers[i]->perf_query_pool;
-               }
+      uint32_t start = 0;
+      uint32_t end = submit->command_buffer_count;
+      struct anv_query_pool *perf_query_pool =
+         cmd_buffers[start]->perf_query_pool;
+      for (uint32_t n = 0; n < end; n++) {
+         bool can_chain = false;
+         uint32_t next = n + 1;
+         /* Can we chain the last buffer into the next one? */
+         if (next < end &&
+             anv_cmd_buffer_is_chainable(cmd_buffers[next]) &&
+             can_chain_query_pools
+             (cmd_buffers[next]->perf_query_pool, perf_query_pool)) {
+            can_chain = true;
+            perf_query_pool =
+               perf_query_pool ? perf_query_pool :
+               cmd_buffers[next]->perf_query_pool;
+         }
+         if (!can_chain) {
+            /* The next buffer cannot be chained, or we have reached the
+             * last buffer, submit what have been chained so far.
+             */
+            VkResult result =
+               anv_queue_exec_locked(queue,
+                                     start == 0 ? submit->wait_count : 0,
+                                     start == 0 ? submit->waits : NULL,
+                                     next - start, &cmd_buffers[start],
+                                     next == end ? submit->signal_count : 0,
+                                     next == end ? submit->signals : NULL,
+                                     perf_query_pool,
+                                     submit->perf_pass_index);
+            if (result != VK_SUCCESS)
+               return result;
+            if (next < end) {
+               start = next;
+               perf_query_pool = cmd_buffers[start]->perf_query_pool;
             }
          }
-
-         VkResult result =
-            anv_queue_exec_locked(queue,
-                                  start == 0 ? submit->wait_count : 0,
-                                  start == 0 ? submit->waits : NULL,
-                                  i - start, &cmd_buffers[start],
-                                  i == end ? submit->signal_count : 0,
-                                  i == end ? submit->signals : NULL,
-                                  perf_query_pool,
-                                  submit->perf_pass_index);
-         if (result != VK_SUCCESS)
-            return result;
-
-         perf_query_pool = NULL;
-         start = i;
       }
-      assert(start == end);
    }
-
    for (uint32_t i = 0; i < submit->signal_count; i++) {
       if (!vk_sync_is_anv_bo_sync(submit->signals[i].sync))
          continue;