STM32F429, external interrupt errata?

[MSipo]

External interrupt errata?

I have setup EXTI_Line12 and EXTI_Line14 to generate en interrupt.

But the EXTI15_10_IRQHandler() fails to execute a second interrupt if it happens too close to the first interrupt.

But it seems only happen when I use bit-banding to check and clear the bits in the EXTI->PR register. If I access the EXTI->PR directly at the normal address it does not happen.

Is this a known problem?

I read something similar in another chip errata (but on an STM8):

When an interrupt handler starts executing a service routine and an external interrupt (EXTI) request is pending or arrives during the same cycle, the external interrupt is not executed.

void EXTI15_10_IRQHandler(void)
{
 GPIO_SetBits(GPIOB, GPIO_Pin_10); // Toggle PB10 to show we are entering the interrupt
 GPIO_ResetBits(GPIOB, GPIO_Pin_10);
 
 // NOTE:
 // PERIPH_GET_BIT is a macro that reads one bit using bit-banding
 // PERIPH_SET_BIT is a macro that sets a bit using bit-banding
 
#ifdef USE_BIT_BANDING
 if (PERIPH_GET_BIT(EXTI->PR, EXTI_Line12))
#else
 if (EXTI->PR & EXTI_Line12)
#endif
 {
 // Clear the EXTI line pending bit
#ifdef USE_BIT_BANDING
 PERIPH_SET_BIT(EXTI->PR, EXTI_Line12);
#else
 EXTI->PR = EXTI_Line12;
#endif
 // Keep PB10 high as long as we do stuff A
 GPIO_SetBits(GPIOB, GPIO_Pin_10);
 do_stuff_a();
 GPIO_ResetBits(GPIOB, GPIO_Pin_10);
 }
 
#ifdef USE_BIT_BANDING
 if (PERIPH_GET_BIT(EXTI->PR, EXTI_Line14))
#else
 if (EXTI->PR & EXTI_Line14)
#endif
 {
 // Clear the EXTI line pending bit
#ifdef USE_BIT_BANDING
 PERIPH_SET_BIT(EXTI->PR, EXTI_Line14);
#else
 EXTI->PR = EXTI_Line14;
#endif
 // Keep PB10 high as long as we do stuff B
 GPIO_SetBits(GPIOB, GPIO_Pin_10); 
 do_stuff_b();
 GPIO_ResetBits(GPIOB, GPIO_Pin_10);
 }
}

[Ozone]

If I remember correctly, there was a race condition problem when you cleared an interrupt flag late in the handler, while another instance happened.

This is AFAIK a core-related issue (ARM), not STM32 related.

[MSipo]

​Yes, I know. I hade that problem in a timer interrupt where I only cleared the interrupt flag and toggle an IO-pin. I needed to insert a dummy read/write access to a variable before leaving the interrupt to let the interrupt flag get time to clear.

[waclawek.jan]

You are not supposed to use a read-modify-write to clear the EXTI->PR bit, you should use a direct write. Bit-bandig writes are internally read-modify-write.

While this writeup is focused on the timer, the point is the same. (Sorry for the formatting, messed up courtesy of ST's forum software migration.)

JW