2

There are several libraries like WiringPi, RPi and pigpio, claiming to implement interrupt handling for GPIO signals. But as far as I can estimate, they all do polling on the pins, therefore implement a busy wait in a parallel thread. Only the library bcm2835 warns that it's not using real interrupts.

Reviewing the available datasheets (BCM2835 and BCM2711) I was able to identify the registers on which one can enable interrupts for three pins, which then should be routed from the Video Core over the Generic Interrupt Controller (GIC-400) to the CPU cores. On ATmega32 you can write to registers like these to enable interrupts on certain pins and implement ISR(desired_vec) as interrupt handler method.

Now I'd like to know how to implement an ISR on RPi. Is that even possible in user space?

And what is the difference between interrupts and BCM's edge detection? Does this actually activate the GIC?

1
  • A good question!
    – Seamus
    Jan 11 at 18:57
3

There is no way to call userspace code from an ISR. Unlike system calls which run on the stack of the userspace program, interrupt handlers use internal kernel memory for the stack. Since that memory is not visible in userspace, the system would crash the moment your ISR userspace function finishes or tries to use the stack for local variables (if not before, due to other reasons I have overlooked).

If you need to play with interrupts, you need to write a kernel driver.

5
  • Thanks for this clear no. Can you advise me any reference on how to handle this on kernel level? Would I have to write to the device tree like here? And havingthis done, I guess I need a bit code that implements my interrupt handling? And actually there was a second question: What exactly is edge detection in comparison?
    – void
    Jan 11 at 19:09
  • AFAIK there's no way to define interrupts from userspace either, OFC every OS under the sun has to deal with interrupts but I understand it's something possible at kernel level only. And sure there are provisions to build things looking like an interrupt.
    – MaxDZ8
    Jan 17 at 15:55
  • @MaxDZ8 "sure there are provisions to build things looking like an interrupt" - yes, that would be poll()/select() and similar functions. Since the kernel cannot call a userspace function, the userspace program has to call the kernel, and returning from the kernel will indicate an event. Jan 18 at 8:23
  • Excuse me @DmitryGrigoryev but I doubt your answers are effective in solving the issue we can observe. There is an ongoing misunderstanding poll/select being interrupt-based ARE interrupts. I think the difference is pretty obvious to us.
    – MaxDZ8
    Jan 19 at 7:20
  • @MaxDZ8 If by "issue" you mean userspace interrupts, then there is no solution. Userspace interrupts are not possible in Linux. Jan 19 at 13:57
2

wiringPi uses interrupts, e.g. with the wiringPiISR function.

pigpio uses interrupts, e.g. with the gpioSetISRFunc function.

lgpio uses interrupts. e.g. with the gGpioSetAlertsFunc function.

None of the above use polling or busy waits. I can only assume you are confused because at a low level they use a Linux function called poll. But this function does not poll the GPIO in the sense you mean.

Linux handles interrupts. As part of its interrupt handling it will eventually schedule one of the above functions.

pigpio can additionally use GPIO polling via DMA which happens to be more accurate and reliable for short (few µs) level changes.

1
  • 1
    Yes, that really confuses me. The manpage on poll says it waits for one of a set of file descriptors to become ready to perform I/O. That's not what I would equalize with an IRQ. I'd expect more like a raise_softirq() on a kernel module; Also e.g. the wiringPi code clearly starts an own threat to handle this poll mechanism
    – void
    Jan 11 at 13:31

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.