diff --git a/src/rtapi/uspace_rtapi_app.cc b/src/rtapi/uspace_rtapi_app.cc
index c42c9f2a6e5..e79ff41b842 100644
--- a/src/rtapi/uspace_rtapi_app.cc
+++ b/src/rtapi/uspace_rtapi_app.cc
@@ -704,7 +704,22 @@ static void configure_memory()
                   "mallopt(M_MMAP_MAX, -1) failed\n");
     }
 #endif
-    char *buf = static_cast<char *>(malloc(PRE_ALLOC_SIZE));
+    /*
+     * The following code seems pointless, but there is a non-observable effect
+     * in the allocation and loop.
+     *
+     * The malloc() is forced to set brk() because mmap() allocation is
+     * disabled in a call to mallopt() above. All touched pages become resident
+     * and locked in the loop because of above mlockall() call (see notes in
+     * mlockall(2)). The mallopt() trim setting prevents the brk() from being
+     * reduced after free(), effectively creating an open space for future
+     * allocations that will not generate page faults.
+     *
+     * The qualifier 'volatile' on the buffer pointer is required because newer
+     * clang would remove the malloc(), for()-loop and free() completely.
+     * Marking 'buf' volatile ensures that the code will remain in place.
+     */
+    volatile char *buf = static_cast<volatile char *>(malloc(PRE_ALLOC_SIZE));
     if (buf == NULL) {
         rtapi_print_msg(RTAPI_MSG_WARN, "malloc(PRE_ALLOC_SIZE) failed\n");
         return;
@@ -717,7 +732,7 @@ static void configure_memory()
              * memory and never given back to the system. */
             buf[i] = 0;
     }
-    free(buf);
+    free((void *)buf);
 }
 
 static int harden_rt()