Re: Hard Real-Time?

[Jaromír Doleček <jaromir.dolecek%gmail.com@localhost>]

Le jeu. 16 avr. 2020 à 18:33, Brad Spencer <brad%anduin.eldar.org@localhost> a écrit :
> Interrupts with gpio and RPI won't be a problem.  You can test some
> behavior in a simple manor with gpioirq(4) and gpiopps(4) uses them.
> The thing I have found that is a little annoying is that interrupts
> appear to be delivered only to CPU0 (at least on the RPI3) as far as I
> can tell and this can pose a problem if you are trying to service lots
> of them.

Yes interrupts are bound to CPU0 by default, but you can bind it to
different CPU via kcpuset_*() and interrupt_distribute().

I don't think it's possible to bind it to several CPUs unfortunately,
but if the device can trigger several interrupts you can bind each to
different CPU.

This is used e.g. in nvme(4) and wm(4).

Jaromir