[Xenomai-help] Switch mode with x86

[Xenomai-help] Switch mode with x86

[Nicolas BLANCHARD]

Hi ! 

I'm newbie in xenomai world,
I've write an application which make a lot of switch mode
(Primary->secondary ...)
and sometime tasks are block in secondary mode (/proc/xenomai/sched ->
stat X).
(no watchod message in dmesg, it's block in secondary mode).

I use xenomai 2.2.5, adeos 1.5.02 (same probleme with 2.2.3 et adeos
1.4)
It' run on x86 (486 vortex 166Mhz).
So, i've write an litle example that failed.  It failed on my x86 (in
less than 2 minutes),
but this example run corectly with a pentium 4 or an Amd processor :

#include <sys/mman.h>
#include <stdio.h>
#include <signal.h>
#include <native/task.h>
#include <string.h>

static int iStop = 0;
RT_TASK Task_1, Task_2;
//----------------------------------------------------------------------THREAD
1
void thread_1(void *arg)
{
	while(1)
	{
		unsigned long overruns;
		int iErr;

		iErr = rt_task_wait_period(&overruns);
		if (iErr || overruns)
			printf("Echec wait_period task_1 err %d,
overruns: %lu\n",iErr,overruns);

		//iErr = rt_task_set_mode(T_PRIMARY, 0, NULL);
		if(iErr)
			printf("Echec set_mod Secondary task_1\n");
printf("r\n");
		if(iStop == 1)
		{
			iErr = rt_task_delete(&Task_1);
			if (iErr ) 
				printf("Echec delete task_1 err %d\n",
iErr);
			break;
		}
	}
}
//----------------------------------------------------------------------THREAD
2
void thread_2(void *arg)
{
	while(1)
	{
		unsigned long overruns;
		int iErr;

		iErr = rt_task_wait_period(&overruns);
		if (iErr || overruns)
			printf("Echec wait_period task_2 err %d,
overruns: %lu\n",iErr,overruns);

		//iErr = rt_task_set_mode(T_PRIMARY, 0, NULL);
		if(iErr)
			printf("Echec set_mod Secondary task_2\n");
printf("u\n");

		if(iStop == 1)
		{
			iErr = rt_task_delete(&Task_2);
			if (iErr ) 
				printf("Echec delete task_2 %d\n",
iErr);
			break;
		}
	}
}
//----------------------------------------------------------------------CLEANUP
UPON SIG
void cleanup_upon_sig(int sig)
{
	iStop = 1;
}
//----------------------------------------------------------------------MAIN
int main(int argc, char *argv[]) 
{
	sigset_t mask, oldmask;
	int iErr;

	mlockall(MCL_CURRENT | MCL_FUTURE);

	sigemptyset(&mask);
	sigaddset(&mask, SIGINT);
	signal(SIGINT, cleanup_upon_sig);
	sigaddset(&mask, SIGTERM);
	signal(SIGTERM, cleanup_upon_sig);
	sigaddset(&mask, SIGHUP);
	signal(SIGHUP, cleanup_upon_sig);

	pthread_sigmask(SIG_BLOCK, &mask, &oldmask);

		// Task 1 *******************************
	iErr = rt_task_create(&Task_1,"TASK1", 0, 99, 0);
	if(iErr)
		printf("Echec create Task_1
%d\n",iErr);			
	iErr = rt_task_set_periodic(
&Task_1,rt_timer_read()+100*1000000llu,rt_timer_ns2ticks(10*1000000llu)
); //period 10 ms
	if(iErr)
		printf("Echec set_periodic Task_1 %d\n",iErr);
	iErr = rt_task_start(&Task_1,(void *)thread_1,NULL);
	if(iErr)
		printf("Echec start Task_1 %d\n", iErr);

		// Task 2 ******************************
	iErr = rt_task_create(&Task_2,"TASK2", 0, 0, 0);
	if(iErr)
		printf("Echec create task_2
%d\n",iErr);			
	iErr = rt_task_set_periodic(
&Task_2,rt_timer_read()+100*1000000llu,rt_timer_ns2ticks(11*1000000llu)
); //period 11 ms
	if(iErr)
		printf("Echec set_periodic task_2 %d\n", iErr);
	iErr = rt_task_start(&Task_2,(void *)thread_2,NULL);
	if(iErr)
		printf("Echec start task_2 %d\n", iErr);

	sigsuspend(&oldmask);
	return 0;
}

If thoses two task have the same priority it looks to not failed,
perhaps a problem
with the scheduler in secondary mode when two task comes from primary
mode ?
Is it a know probleme with x86 ?
Or perhaps a big error in my example ...

Have you an idea ? 
thanks in advance

nicolas blanchard

Re: [Xenomai-help] Switch mode with x86

[Jan Kiszka]

[-- Attachment #1: Type: text/plain, Size: 2739 bytes --]

Nicolas BLANCHARD wrote:
> Hi ! 
> 
> I'm newbie in xenomai world,
> I've write an application which make a lot of switch mode
> (Primary->secondary ...)
> and sometime tasks are block in secondary mode (/proc/xenomai/sched ->
> stat X).
> (no watchod message in dmesg, it's block in secondary mode).
> 
> I use xenomai 2.2.5, adeos 1.5.02 (same probleme with 2.2.3 et adeos
> 1.4)
> It' run on x86 (486 vortex 166Mhz).
> So, i've write an litle example that failed.  It failed on my x86 (in
> less than 2 minutes),
> but this example run corectly with a pentium 4 or an Amd processor :

How does it fail precisely? Which tasks are stuck? Attach
/proc/xenomai/sched output e.g. Already tried to attach gdb to that
process after the lock-up?


One difference of vortex vs. P4 might be (besides speed) that for the
former Xenomai uses TSC emulation, for the latter the CPU already comes
with this support. You can force the high-end systems into that mode by
compiling the kernel with CONFIG_M586. Or did you use the same kernel
images on all systems?

Your .config would be interesting. You can already check if
CONFIG_INPUT_PCSPKR is set. Must be off for TSC emulation (Xenomai 2.2
doesn't check for this conflict, 2.3 will do so).

> 
> #include <sys/mman.h>
> #include <stdio.h>
> #include <signal.h>
> #include <native/task.h>
> #include <string.h>
> 
> static int iStop = 0;
> RT_TASK Task_1, Task_2;
> //----------------------------------------------------------------------THREAD
> 1
> void thread_1(void *arg)
> {
> 	while(1)
> 	{
> 		unsigned long overruns;
> 		int iErr;
> 
> 		iErr = rt_task_wait_period(&overruns);
> 		if (iErr || overruns)
> 			printf("Echec wait_period task_1 err %d,
> overruns: %lu\n",iErr,overruns);

Would have been better to attach the program instead of inlining it. :)

> 
> 		//iErr = rt_task_set_mode(T_PRIMARY, 0, NULL);
> 		if(iErr)
> 			printf("Echec set_mod Secondary task_1\n");
> printf("r\n");
> 		if(iStop == 1)
> 		{
> 			iErr = rt_task_delete(&Task_1);
> 			if (iErr ) 
> 				printf("Echec delete task_1 err %d\n",
> iErr);
> 			break;
> 		}
> 	}

There is no mode switch in this thread and the second one. Is this not
required to cause the lock-up???

...
> 
> If thoses two task have the same priority it looks to not failed,
> perhaps a problem
> with the scheduler in secondary mode when two task comes from primary
> mode ?
> Is it a know probleme with x86 ?

Nope. Fairly unusual. Does the switchtest from the Xenomai test-suite
fail as well?

> Or perhaps a big error in my example ...

Nothing I can see.

> 
> Have you an idea ? 
> thanks in advance
> 
> nicolas blanchard
> 

Jan


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 250 bytes --]