Re: Relax tracing not working for me

[Philippe Gerum <rpm@xenomai.org>]

"Rosenow, James" <jim.rosenow@mts.com> writes:

> I am using Xenomai v3.2.1 on 5.15.57-dovetail and I have selected the following kernel configuration option.
>
> .config - Linux/x86 5.15.57 Kernel Configuration
>  Xenomai/cobalt > Debug support
> [*]   Trace relax requests
>
> When I run my application, I am getting MSW's on one of my real-time tasks, i.e. seqTask-00 (below).
>
> Every 10.0s: cat /proc/xenomai/sched/stat                                 test: Tue Jan  7 16:32:14 2025
> CPU  PID    MSW        CSW        XSC        PF    STAT       %CPU  NAME
>   0  0      0          3526511    0          0     00018000   99.8  [ROOT/0]
>   1  0      0          22166547   0          0     00018000   98.7  [ROOT/1]
>   2  0      0          6          0          0     00018000  100.0  [ROOT/2]
>   3  0      0          0          0          0     00018000  100.0  [ROOT/3]
>   4  0      0          12         0          0     00018000  100.0  [ROOT/4]
>   5  0      0          0          0          0     00018000  100.0  [ROOT/5]
>   0  311    0          2          0          0     00000042    0.0  [rtnet-stack]
>   0  312    0          2          0          0     00020042    0.0  [rtnet-rtpc]
>   0  34201  8          8          54         0     000480c0    0.0  coops_target_main
>   0  34203  714420     714421     1429756    0     000480c0    0.2  userCycle
>   1  34204  2          2          571184     0     000480c0    0.0  clockInterrupt
>   1  34205  1          571183     1356559    0     00048042    0.4  dispatch
>   1  34206  571182     1142364    713982     0     00048042    0.5  seqTask-00
>   1  34207  1          285594     356993     0     00048042    0.3  seqTask-01
>   1  34208  1          142798     180963     0     00048042    0.1  seqTask-02
>   1  34209  1          2          3          0     00048042    0.0  seqTask-03
>   1  34210  1          2          3          0     00048042    0.0  seqTask-04
>
> This is an iterative task that is executing at 2000Hz, I get 1 additional MSW for each run iteration.  As you can see there are many MSW's.  Shouldn't the trace buffer be full of these relaxations?  If I cat the buffer, I see nothing.
>
> test@test:~$ cat /proc/xenomai/debug/relax
> test@test:~$
>
> What am I doing wrong?
>

In short, you may want to make sure that PTHREAD_WARNSW is set for
'seqTask-00'. The STAT field for that task reveals that it's not (0x4000
is clear there).

Why? Because the logic of MSW tracing is as follows when
CONFIG_XENO_OPT_DEBUG_TRACE_RELAX is set:

1. On MSW detected by the core for the current thread _and only if_
PTHREAD_WARNSW is set for such thread (which is running a syscall), the
latter is immediately sent SIGWINCH(aka SIGSHADOW), with
info=SIGSHADOW_ACTION_BACKTRACE (xnthread_relax ->
xndebug_notify_relax).

2. The pre-defined SIGSHADOW handler in libcobalt.so (see
lib/cobalt/sigshadow.c) runs, taking a snapshot of the current stack
backtrace, then sends it to the core via Cobalt's 'backtrace' system
call. Since we are not using alternate stacks, the code which caused the
MSW must appear somewhere in the lower frames below the current signal
frame. This location is figured out by the slackspot utility when
analyzing the captured frames.

3. The core receives the backtrace data, then logs it into the
appropriate trace buffer which is behind /proc/xenomai/debug/relax
(backtrace -> xndebug_trace_relax, cobalt/posix/syscall.c).

HTH,

-- 
Philippe.