August firmware breaks nanosleep

[daves]

The timing/sleeping method advised to be used in C loops etc seems to have broken in the latest firmware.

 

After a lot of investigation into freezing downloads and failed reboots I have narrowed the problem down to a repeatable example:

 

int main(void) 
{
struct timespec SecSleeper = {0};	// Initialize time structure
SecSleeper.tv_sec = 1;			// 1e9 nanoseconds, which equals 1 sec
SecSleeper.tv_nsec = 0;			// 1e9 nanoseconds, which equals 1 sec

InitLibrary();

pshm->P[1] = 0;

// Main loop just keeps us running
while (1)
{
	printf("Dummy\n");
	pshm->P[1]++;
	pshm->P[2] = nanosleep(&SecSleeper, NULL);
	pshm->P[3] = errno;
}

CloseLibrary();	
return 0;
}

 

Run capp1.out on v1.4.0.27 from a putty terminal and you see "Dummy" output once a second, looking in watch window shows P1 incrementing once a second. The return value P2=0 (success) and errno P3=0 (no error)

 

Run capp1.out on v1.5.3.0 from putty terminal and you see "Dummy" output constantly, looking in watch window shows P1 incrementing almost 100,000 time a second. The return value P2=-1 (failure) and errno P3=1 (EPERM, operation not permitted)

 

This looks like a Xenomai error saying the calls are coming from an invalid context.

 

 

Reverting to 1.4.027 fixes the problem. I will have to do this as I have many C apps and CPLCs which rely on this method. Or is there a new method for timing in PPMAC?

 

Can you confirm you see the same issue?

 

The August IDE complains about the firmware now so please can I have instructions on how to downgrade the IDE cleanly?

[shansen]

If nanosleep returns EPERM that suggests that Xenomai thinks it is not in a real-time context. Are you sure that you have called pthread_setschedparam on the current thread?

[hbausley]

The timing/sleeping method advised to be used in C loops etc seems to have broken in the latest firmware.

 

After a lot of investigation into freezing downloads and failed reboots I have narrowed the problem down to a repeatable example:

 

int main(void) 
{
struct timespec SecSleeper = {0};	// Initialize time structure
SecSleeper.tv_sec = 1;			// 1e9 nanoseconds, which equals 1 sec
SecSleeper.tv_nsec = 0;			// 1e9 nanoseconds, which equals 1 sec

InitLibrary();

pshm->P[1] = 0;

// Main loop just keeps us running
while (1)
{
	printf("Dummy\n");
	pshm->P[1]++;
	pshm->P[2] = nanosleep(&SecSleeper, NULL);
	pshm->P[3] = errno;
}

CloseLibrary();	
return 0;
}

 

Run capp1.out on v1.4.0.27 from a putty terminal and you see "Dummy" output once a second, looking in watch window shows P1 incrementing once a second. The return value P2=0 (success) and errno P3=0 (no error)

 

Run capp1.out on v1.5.3.0 from putty terminal and you see "Dummy" output constantly, looking in watch window shows P1 incrementing almost 100,000 time a second. The return value P2=-1 (failure) and errno P3=1 (EPERM, operation not permitted)

 

This looks like a Xenomai error saying the calls are coming from an invalid context.

 

 

Reverting to 1.4.027 fixes the problem. I will have to do this as I have many C apps and CPLCs which rely on this method. Or is there a new method for timing in PPMAC?

 

Can you confirm you see the same issue?

 

The August IDE complains about the firmware now so please can I have instructions on how to downgrade the IDE cleanly?

 

 

SHansen was on the right track. We added the call below to your example and it sleeps.

param.__sched_priority = 1;

pthread_setschedparam(pthread_self(), SCHED_FIFO, &param);

 

nanosleep is reserved to xenomai threads. The rtos was compiled without the option that causes the main thread to be a xenomai thread. We recompiled the rtos with this option. We will post a powerpmac.deb version 1.5.4.0 with that option restored so it works without pthread_setschedparam as it did previously.

 

As a rule for realtime code use nanosleep and for generic non realtime code use sleep or usleep.

[sponseld]

When can we expect that the powerpmac.deb version 1.5.4.0 with that option restored will be available? We also use nanosleep and hope to avoid having to put in the pthread_setschedparam fix. It has been over a month since the last post above, but the latest version in the File Depot is still 1.5.3.0.

[hbausley]

When can we expect that the powerpmac.deb version 1.5.4.0 with that option restored will be available? We also use nanosleep and hope to avoid having to put in the pthread_setschedparam fix. It has been over a month since the last post above, but the latest version in the File Depot is still 1.5.3.0.

 

We will have a new version posted within a week.