Index: uspace/app/perf/doc/doxygroups.h
===================================================================
--- uspace/app/perf/doc/doxygroups.h	(revision 60029dfe5895b620ab28744ecb20cab6de3b9b8b)
+++ uspace/app/perf/doc/doxygroups.h	(revision a362c166d0d57f9e8ef9b820030c62e68272a36e)
@@ -2,3 +2,41 @@
  * @brief User space performance measuring tool
  * @ingroup apps
+ *
+ * @details
+ *
+ * To add a new benchmark, you need to implement the actual benchmarking
+ * code and register it.
+ *
+ * Registration is done by adding
+ * <code>extern benchmark_t bench_YOUR_NAME</code> reference to benchlist.h
+ * and by adding it to the array in benchlist.c.
+ *
+ * The actual benchmark should reside in a separate file (see malloc/malloc1.c
+ * for example) and has to (at least) declare one function (the actual
+ * benchmark) and fill-in the benchmark_t structure.
+ *
+ * Fill-in the name of the benchmark, its description and a reference to the
+ * benchmark function to the benchmark_t.
+ *
+ * The benchmarking function has to accept for arguments:
+ *  @li stopwatch_t: store the measured data there
+ *  @li uint64_t: size of the workload - typically number of inner loops in
+ *      your benchmark (used to self-calibrate benchmark size)
+ *  @li char * and size_t giving you access to buffer for storing error message
+ *  if the benchmark fails (return false from the function itself then)
+ *
+ * Typically, the structure of the function is following:
+ * @code{c}
+ * static bool runner(stopwatch_t *stopwatch, uint64_t size,
+ *     char *error, size_t error_size)
+ * {
+ * 	stopwatch_start(stopwatch);
+ * 	for (uint64_t i = 0; i < size; i++) {
+ * 		// measured action
+ * 	}
+ * 	stopwatch_stop(stopwatch);
+ *
+ * 	return true;
+ * }
+ * @endcode
  */