Tickless idle atmel sam3x device

wlauer wrote on Thursday, November 20, 2014:

I’ve been experimenting with the tickless idle functionality and have a few questions. I need to put our device into a lower power mode which involves changing the main clock speed to a lower value and restoring it upon waking. I’ve figured out all the details and it works as expected.

But I ran into an issue where the device would enter sleep mode while waiting on peripherals to complete. Unfortunately when entering sleep the resulting change in the main clock also changes affects the peripherals operation. So I modified the sleepmgr code that atmel provides to provide a locking mechanism around the code that deals with the peripherals.

As you can see I wrapped a mutex around a variable array that keeps the lock count for various depths of sleeping. So I though I would put the call sleepmgr_get_sleep_mode() to determine if we should sleep or not into the portSUPPRESS_TICKS_AND_SLEEP call only to realize that portSUPPRESS_TICKS_AND_SLEEP is just another hook into the Idle task which can have no blocking calls.

How should I be handling this? How do i get around not blocking in the idle task?

Thanks in advance,
Bill

a snippet of the sleepmgr code:

static inline void sleepmgr_init(void)
{
uint8_t i;

/* semaphore used to access the sleep level as it is written to from more than one task. */
sleepmgr_mutex = xSemaphoreCreateMutex();
xSemaphoreTake(sleepmgr_mutex, portMAX_DELAY);

for (i = 0; i < SLEEPMGR_NR_OF_MODES - 1; i++) {
	sleepmgr_locks[i] = 0;
}
sleepmgr_locks[SLEEPMGR_NR_OF_MODES - 1] = 1;
xSemaphoreGive(sleepmgr_mutex);

}

static inline void sleepmgr_lock_mode(enum sleepmgr_mode mode)
{
Assert(sleepmgr_locks[mode] < 0xff);
xSemaphoreTake(sleepmgr_mutex, portMAX_DELAY);
++sleepmgr_locks[mode];
xSemaphoreGive(sleepmgr_mutex);
}

static inline enum sleepmgr_mode sleepmgr_get_sleep_mode(void)
{
xSemaphoreTake(sleepmgr_mutex, portMAX_DELAY);
enum sleepmgr_mode sleep_mode = SLEEPMGR_ACTIVE;

uint8_t *lock_ptr = sleepmgr_locks;
// Find first non-zero lock count, starting with the shallowest modes.
while (!(*lock_ptr)) {
	lock_ptr++;
	sleep_mode++;
}

// Catch the case where one too many sleepmgr_unlock_mode() call has been
// performed on the deepest sleep mode.
Assert((uintptr_t)(lock_ptr - sleepmgr_locks) < SLEEPMGR_NR_OF_MODES);
xSemaphoreGive(sleepmgr_mutex);

return sleep_mode;

}

davedoors wrote on Friday, November 21, 2014:

The semaphore is probably only needed when changing the lock count. So tasks that update the lock count could use the functions with the mutex, and the idle task could use a function without a mutex just to read the lock counts as long is it does not change them. Would that work for you?