[En-Nut-Discussion] Interrupts on at91 plattform: Edge vs. level triggered

[Thiago A. Corrêa]

Hi Ole,

   This might be completely unrelated, but while I'm trying to figure
out what to do and where with the AVR32, I'm looking at the arm code
and I just noticed something:

#else /* NUT_CRITICAL_NESTING_STACK */

extern unsigned int critical_nesting_level;

#define NutEnterCritical() \
if (critical_nesting_level++ == 0) { \
    int temp_; \
    asm volatile (             \
            "@ NutEnterCritical"    "\n\t" \
            "mrs     %0, cpsr"      "\n\t" \
            "orr     %0, %0, #0xC0" "\n\t" \
            "msr     cpsr, %0"      "\n\t" \
            : "=r" (temp_) : : "cc"); \
}

Shouldn't we disable interrupts first then change the variable? It
looks unsafe the way it is implemented

#define NutExitCritical() \
if (critical_nesting_level) { \
    int temp_; \
    asm volatile (             \
            "@ NutExitCritical"     "\n\t" \
            "mrs     %0, cpsr"      "\n\t" \
            "bic     %0, %0, #0xC0" "\n\t" \
            "msr     cpsr, %0"      "\n\t" \
            : "=r" (temp_) : : "cc"); \
    critical_nesting_level--; \
}

Same here, I think the variable update should be done before enabling
interrupts.

What do you think?

Kind Regards,
    Thiago A. Correa

On Tue, Feb 3, 2009 at 3:19 PM, Ole Reinhardt
<ole.reinhardt at embedded-it.de> wrote:
> Hey!
>
> I just noticed several times problems with drivers using edge triggered
> interrupts on the arm platform. At least you remember the UART bug,
> fixed at 2008-07-26 by Harald (see Changelog).
>
> I noticed similar problems with the GPIO level change interrupts as
> well. If I leave the interrupts in their default trigger mode (edge
> triggered) the interrupt handler will not be triggered again after some
> thousand interrupts where generated before. It seems that this might
> happen if the Interrupts are disabled by a critical section just in the
> moment when the interrupt occurs. It seems that the interrupt flag is
> set in the meantime and therefore will never change again until it is
> cleared. But it will only be cleared in the interrupt handler which was
> not called as there was no "change" detected when interrupts where
> enabled.
>
> So two questions:
>
> - Is there any reason to use edge triggered interrupts instead of level
>  triggered for the internal peripherals?
> - If yes: Is there any way to workaround this problem? For example by
>  checking if an interrupt should have been triggered when leaving the
>  critical section?
>
> We should carefully check all existing drivers if this problem might
> occur there too. Especialy this could solve the problems with the TWI
> driver too (not yet checked).
>
> Regards,
>
> Ole Reinhardt
>
> --
>  _____________________________________________________________
> |                                                             |
> | Embedded-IT          Hard- und Softwarelösungen             |
> |                                                             |
> | Ole Reinhardt        Tel. / Fax:        +49 (0)271  7420433 |
> | Luisenstraße 29      Mobil:             +49 (0)177  7420433 |
> | 57076 Siegen         eMail:    ole.reinhardt at embedded-it.de |
> | Germany              Web:         http://www.embedded-it.de |
> |                      UstID / VAT:       DE198944716         |
> |_____________________________________________________________|
>
> _______________________________________________
> http://lists.egnite.de/mailman/listinfo/en-nut-discussion
>