HAL os-less & ISR execution issue Topic is solved

[eblot]

I'm trying to use the HAL (os-less) mode for a PoC project (STM32L0, NUCLEO64_L053R8 and NUCLEO32_L011K4).

I've enabled one serial port (SD2) for debugging traces, one timer (TIM2), and a couple of GPIO.
STM32 uses HSI16 as the main clock source. I can observe the (scaled) clock on a GPIO (using MCO)

I'm running TIM2 with HAL_GPT in continuous mode, with a callback (4MHz, ARR: 0xFFFF).
The callback is configured to toggle a GPIO on each call.

I can observe that the callback never gets called when the main() function is running a busy loop (something like for(;;) {}).
If I add some call to osalThreadSleepMilliseconds() in this loop, the callback is called.

I would have expected that the "main" (unique) thread would get preempted by the TIM2 ISR, but it seems it is not, so I guess I missed an important point somewhere (configuration)? I'm pretty sure it is not related to TIM2, as I can observe that the SD2 traces do not get output if a busy loop is called right after chprintf() has been called - I guess same symptom, ISR is not executed.

There is something critical I did not get about ChibiOS & OS-less mode it seems. I would have expected the ISR to execute even if the main() function was running a busy loop.

Any clue?
I tried to locate some demo about os-less environments, but I failed to find a matching example.

I've read that osalThreadSleepMilliseconds() uses a virtual timer, which seems to be documented as a part of the RT subsystem, so I got even more confused

Unrelated question: is the IRQ priority level (order, boundaries, ..) documented somewhere?

Thanks

[Giovanni]

Hi,

Do you enable interrupts in your main() ? the main function is entered with interrupts disabled.

Giovanni

[eblot]

I call
osalSysEnable();
once HW configuration in complete (before the busy loop), which call __enable_irq() if I'm not mistaken.

I just give a tried and force OSAL_ST_MODE == OSAL_ST_MODE_NONE and it seems the ISR handler is called as expected. I'm not sure why.

As this setting is not part of the osalconf.h, I never tried to changed it up to now - and I may be on the wrong way here.
I guess that w/o it, I will not be able to call functions such as osalThreadSleepMilliseconds() as there is no tick to wake the thread up.

[eblot]

If I use the following kludge in the busy loop:

Code: Select all

   for(;;) {
      __enable_irq();
      __disable_irq();
   }

the TIM2 ISR handler runs.

If I only move the IRQ enablement before the loop:

Code: Select all

    __enable_irq();
   for(;;) {
   }

the TIM2 ISR does not get called anymore.

I guess that "something" fails to restore the IRQ, but I do not know (yet) what.
The only callback I used is dedicated to TIM2 and is rather simple:

Code: Select all

static void
_ow_gpt_cb(GPTDriver * gptp) {
   (void)gptp;
   _ow_ge.ge_ticks += 1;
   palTogglePad(LED1_GPIO_PORT, LED1_GPIO_PIN);
}

static const GPTConfig _GPT2_CONFIG = {
   .frequency = STM32_HSI16CLK >> 2U,
   .callback = _ow_gpt_cb,
};

struct ow_gpt_engine {
   volatile uint32_t ge_ticks;
};
struct ow_gpt_engine _ow_ge;

[eblot]

From ChibiOS_18.2.1/os/hal/osal/os-less/ARMCMx/osal.h

Code: Select all

/**
 * @brief   Enters a critical zone from ISR context.
 * @note    This function cannot be used for reentrant critical zones.
 *
 * @special
 */
static inline void osalSysLockFromISR(void) {

#if CORTEX_MODEL == 0
  __disable_irq();
#else
  __set_BASEPRI(OSAL_IRQ_MAXIMUM_PRIORITY);
#endif
}

/**
 * @brief   Leaves a critical zone from ISR context.
 * @note    This function cannot be used for reentrant critical zones.
 *
 * @special
 */
static inline void osalSysUnlockFromISR(void) {
   // ?
}

I would have expected

Code: Select all

osalSysUnlockFromISR()

to be symetrical with

Code: Select all

osalSysLockFromISR()

and call

Code: Select all

__enable_irq()

Is it on purpose?

[Giovanni]

Hi,

It is the error I think, modify as follow:

Code: Select all

static inline void osalSysUnlockFromISR(void) {

#if CORTEX_MODEL == 0
  __enable_irq();
#else
  __set_BASEPRI(0);
#endif
}

Moving this topic in bug reports, it will be fixed in next release.

Giovanni

[eblot]

Hi Giovanni,

I changed the function with this code (when I posted the question actually), and it indeed fixed the problem, but I was not sure it was the root cause.

Thanks for your help,
Manu.

While seeking the code, I noticed a couple of typos in the comments, unrelated to this issue.
I'm not sure to report this (pull request?), sorry for this off-topic add-on:

* os/hal/ports/STM32/LLD/TIMv1/hal_st_lld.c@248:

Code: Select all

 * @brief   TIM2 interrupt handler.

-> free running counter may be different than TIM2

* os/common/ext/ST/STM32L0xx/stm32*.h

Code: Select all

#define RCC_CFGR_MCO_PRE_2                  RCC_CFGR_MCOPRE_DIV2        /*!< MCO is divided by 1 */
#define RCC_CFGR_MCO_PRE_4                  RCC_CFGR_MCOPRE_DIV4        /*!< MCO is divided by 1 */
#define RCC_CFGR_MCO_PRE_8                  RCC_CFGR_MCOPRE_DIV8        /*!< MCO is divided by 1 */
#define RCC_CFGR_MCO_PRE_16                 RCC_CFGR_MCOPRE_DIV16       /*!< MCO is divided by 1 */

-> divider value comment is not accurate.

[Giovanni]

Fixed as bug #987.

Giovanni