Re: [PATCH] tracing: Have trace_marker writes be just half of TRACE_SEQ_SIZE

[Steven Rostedt <rostedt@goodmis.org>]

On Mon, 4 Mar 2024 20:36:28 -0500
Mathieu Desnoyers <mathieu.desnoyers@efficios.com> wrote:

> >             <...>-999     [001] .....  2296.140373: tracing_mark_write: hello
> > ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
> >     This is the meta data that is added to trace_seq  
> 
> If this header has a known well-defined upper-limit length, then use
> that in the BUILD_BUG_ON().

Unfortunately there's no set limit. It's built up by different callbacks
and such. The output can be changed by options set by the user and even by
tracers, like the function graph tracer:

# tracer: function_graph
#
# CPU  DURATION                  FUNCTION CALLS
# |     |   |                     |   |   |   |
 1)               |  /* hello */


But the worse that will happen if it overflows is that the event is
replaced with:

             <...>-999     [001] .....  2296.140373: [LINE TOO BIG]

But this has never happened outside of development. I guess you could
trigger it if you add a trace_printk() that has a string bigger than
TRACE_SEQ_BUFFER_SIZE. But as Linus says, "Don't do stupid things" ;-)

But in reality, with all the options and everything, I've never seen the
appended text more than 80 bytes (and probably much less).

-- Steve