[Xenomai-help] Problem with mutexes in daemon

[Xenomai-help] Problem with mutexes in daemon

[Roman Pisl]

[-- Attachment #1: Type: text/plain, Size: 419 bytes --]

Hello,
I'm experiencing weird behavior with mutexes in a daemon application. 
After daemon() is called, unnamed mutexes fail on call 
rt_mutex_release() with -EPERM. Everything is ok for named mutexes.

Is there something special with daemon and unnamed mutexes - should all 
have name? Or is it a bug? I'm using kernel 2.6.28 and Xenomai 2.4.6.1.

Example showing this problem is attached.

Thank you for help,

Roman

[-- Attachment #2: daemon.c --]
[-- Type: text/plain, Size: 3313 bytes --]

#include <stdio.h>
#include <stdbool.h>
#include <signal.h>
#include <unistd.h>
#include <sys/mman.h>

#include <native/task.h>
#include <native/cond.h>
#include <native/mutex.h>
#include <native/timer.h>

RT_TASK demo_task;
RT_TASK main_task;
RT_MUTEX mutex;
RT_MUTEX shut_mutex;
RT_COND shut_cond;

#define TRUE 1
#define FALSE 0

bool terminated = FALSE;
bool terminated2 = FALSE;

void demo(void *arg)
{
    int ret;
    RTIME now, previous;

    rt_task_set_periodic(NULL, TM_NOW, 1000000000);
    previous = rt_timer_read();

    while (!terminated2) {
        rt_task_wait_period(NULL);
        now = rt_timer_read();

        printf("DT: Time since last turn: %ld.%06ld ms\n",
               (long)(now - previous) / 1000000,
               (long)(now - previous) % 1000000);
               previous = now;

        if (ret = rt_mutex_acquire(&mutex, TM_INFINITE))
            printf("DT: rt_mutex_acquire() failed %i\n", ret);
        rt_task_sleep(1000000);
        if (ret = rt_mutex_release(&mutex))
            printf("DT: rt_mutex_release() failed %i\n", ret);
    }
    if (ret = rt_mutex_acquire(&shut_mutex, TM_INFINITE))
        printf("DT: rt_mutex_acquire() before cond failed %i\n", ret);
    if (ret = rt_cond_signal(&shut_cond))
        printf("DT: rt_cond_signal() failed %i\n", ret);
    if (ret = rt_mutex_release(&shut_mutex))
        printf("DT: rt_mutex_release() after cond failed %i\n", ret);
}

void catch_signal(int sig)
{
    terminated = TRUE;
}

int main(int argc, char* argv[])
{
    int ret;
    //when daemon - problems with unnamed mutexes
    //without daemon everything ok
    daemon(1, 1);

    if (ret = rt_mutex_create(&mutex, NULL))
    {
        printf("Error creating mutex %i\n", ret);
        return 0;
    }

    if (ret = rt_mutex_create(&shut_mutex, NULL))
    {
        printf("Error creating shut_mutex %i\n", ret);
        return 0;
    }

    if (ret = rt_cond_create(&shut_cond, NULL))
    {
        printf("Error creating cond%i\n", ret);
        return 0;
    }

    signal(SIGTERM, catch_signal);
    signal(SIGINT, catch_signal);

    mlockall(MCL_CURRENT|MCL_FUTURE);

    if (ret = rt_task_shadow(&main_task, "main", 90, 0))
        printf("rt_task_shadow() failed %i\n", ret);
    if (ret = rt_task_create(&demo_task, "trivial", 0, 99, 0))
        printf("rt_task_create() failed %i\n", ret);

    rt_task_start(&demo_task, &demo, NULL);

    while (!terminated)
    {
        if (ret = rt_mutex_acquire(&mutex, TM_INFINITE))
            printf("MT: rt_mutex_acquire() failed, %i\n", ret);
        rt_task_sleep(1000000);
        printf("sleep\n");
        if (ret = rt_mutex_release(&mutex))
            printf("MT: rt_mutex_release() failed %i\n", ret);
        rt_task_sleep(1000000000);
    }
    printf("Finished...\n");

    if (ret = rt_mutex_acquire(&shut_mutex, TM_INFINITE))
        printf("MT: rt_mutex_acquire() before cond failed %i\n", ret);
    terminated2 = TRUE;
    if (ret = rt_cond_wait(&shut_cond, &shut_mutex, TM_INFINITE))
        printf("MT: rt_cond_signal() failed %i\n", ret);
    if (ret = rt_mutex_release(&shut_mutex))
        printf("MT: rt_mutex_release() after cond failed %i\n", ret);

    rt_task_delete(&demo_task);
    rt_mutex_delete(&mutex);
    rt_mutex_delete(&shut_mutex);
    rt_cond_delete(&shut_cond);
}

Re: [Xenomai-help] Problem with mutexes in daemon

[Gilles Chanteperdrix]

Roman Pisl wrote:
> Hello,
> I'm experiencing weird behavior with mutexes in a daemon application. 
> After daemon() is called, unnamed mutexes fail on call 
> rt_mutex_release() with -EPERM. Everything is ok for named mutexes.
> 
> Is there something special with daemon and unnamed mutexes - should all 
> have name? Or is it a bug? I'm using kernel 2.6.28 and Xenomai 2.4.6.1.
> 
> Example showing this problem is attached.

Unnamed mutexes are local to a process. Since forking results in a
different process, mutexes are not preserved.

Yes, Xenomai behaviour is not intuitive, but fixing this requires
internal changes.

-- 
                                                 Gilles.

Re: [Xenomai-help] Problem with mutexes in daemon

[Roman Pisl]

Gilles Chanteperdrix napsal(a):
> Roman Pisl wrote:
>> Hello,
>> I'm experiencing weird behavior with mutexes in a daemon application. 
>> After daemon() is called, unnamed mutexes fail on call 
>> rt_mutex_release() with -EPERM. Everything is ok for named mutexes.
>>
>> Is there something special with daemon and unnamed mutexes - should all 
>> have name? Or is it a bug? I'm using kernel 2.6.28 and Xenomai 2.4.6.1.
>>
>> Example showing this problem is attached.
> 
> Unnamed mutexes are local to a process. Since forking results in a
> different process, mutexes are not preserved.
> 
> Yes, Xenomai behaviour is not intuitive, but fixing this requires
> internal changes.
> 

I understand, but daemon() is the first call and everything is done in 
the new forked process.

What is the best solution for now? Is everything except unnamed mutexes 
supposed to work well in daemon or should I cancel the concept of daemon 
at all?

Roman

Re: [Xenomai-help] Problem with mutexes in daemon

[Gilles Chanteperdrix]

Roman Pisl wrote:
> Gilles Chanteperdrix napsal(a):
>> Roman Pisl wrote:
>>> Hello,
>>> I'm experiencing weird behavior with mutexes in a daemon application. 
>>> After daemon() is called, unnamed mutexes fail on call 
>>> rt_mutex_release() with -EPERM. Everything is ok for named mutexes.
>>>
>>> Is there something special with daemon and unnamed mutexes - should all 
>>> have name? Or is it a bug? I'm using kernel 2.6.28 and Xenomai 2.4.6.1.
>>>
>>> Example showing this problem is attached.
>> Unnamed mutexes are local to a process. Since forking results in a
>> different process, mutexes are not preserved.
>>
>> Yes, Xenomai behaviour is not intuitive, but fixing this requires
>> internal changes.
>>
> 
> I understand, but daemon() is the first call and everything is done in 
> the new forked process.
> 
> What is the best solution for now? Is everything except unnamed mutexes 
> supposed to work well in daemon or should I cancel the concept of daemon 
> at all?

Yes, what you are doing is supposed to work, I did not look at your
code, I thought you were expecting to be able to create the mutex in a
process and use it in its child.

On what platform do you get these issues ?

-- 
                                                 Gilles.