Re: [PATCH v4 2/4] ftrace: Add support for function argument to graph tracer

[Steven Rostedt]

On Fri, 11 Apr 2025 14:00:40 +0100
Mark Brown <broonie@xxxxxxxxxx> wrote:

> On Thu, Apr 10, 2025 at 01:17:45PM -0400, Steven Rostedt wrote:
> > Mark Brown <broonie@xxxxxxxxxx> wrote:  
> 
> > > We've been seeing the PID filters selftest failing for a while on
> > > several arm64 systems, a bisect I managed to run without running into
> > > any confounding issues pointed to this patch which is in mainline as
> > > ff5c9c576e75.  It's in the ftrace code, but I'm not immediately seeing
> > > the relevance.  Output from a failing run:  
> 
> > Hmm, I wonder if there's junk being added into the trace.  
> 
> > Can you add this patch, and show me the output when it fails again?  
> 

Can you show the information before this output, to see what it is actually
testing?

> # # + cat trace
> # # # tracer: function_graph
> # # #
> # # # CPU  TASK/PID         DURATION                  FUNCTION CALLS
> # # # |     |    |           |   |                     |   |   |   |
> # # 0)  ftracet-5190  | ! 537.633 us  |  kernel_clone(); /* ret=0x1470 */
> # # 
> # # 0)  ftracet-5190  | ! 508.253 us  |  kernel_clone(); /* ret=0x1471 */
> # # 
> # # 0)  ftracet-5190  | ! 215.716 us  |  kernel_clone(); /* ret=0x1476 */
> # # 
> # # 0)  ftracet-5190  | ! 493.890 us  |  kernel_clone(); /* ret=0x147b */
> # # 
> # # + fail PID filtering not working?
> 
> ...
> 
> # # + cat trace
> # # # tracer: function_graph
> # # #
> # # # CPU  TASK/PID         DURATION                  FUNCTION CALLS
> # # # |     |    |           |   |                     |   |   |   |
> # # 0) ftracet-12279  | ! 598.118 us  |  kernel_clone(); /* ret=0x301f */
> # # 
> # # 0) ftracet-12279  | ! 492.539 us  |  kernel_clone(); /* ret=0x3020 */
> # # 
> # # 0) ftracet-12279  | ! 231.104 us  |  kernel_clone(); /* ret=0x3025 */
> # # 
> # # 0) ftracet-12279  | ! 555.566 us  |  kernel_clone(); /* ret=0x302a */
> # # 
> # # + fail PID filtering not working?

Also, is it possible to just enable function_graph tarcing and see if it
adds these blank lines between events?

-- Steve