Non-Maskable Interrupt

[pphaneuf]

I mentioned NMI's (non-maskable interrupt) today and was asked for clarification. I was more accurately referring to a priority interrupt. It has to be noted that during processing of an interrupt, interrupts are normally blocked, to avoid incredible confusion.

This includes the timer interrupt, commonly used for the management of multitasking features in modern operating systems. This is why when too much time is spent in interrupt processing, normal functioning is visibly disrupted (cannot move the mouse cursor, everything is frozen).

While this behaviour might be annoying on an more conventional interactive system, such as a workstation, in some realtime applications this can be a desired behaviour, as the interrupt processing often takes care of a critical functionality of the system (an example would be the audio decoding of an iPod).

Hoping that this clears things up, I shall go to sleep now.

Comments

[azrhey.livejournal.com]

call me ignorant, but as I read your post, one questions comes to mind, isn't that what backup systems are for?

*confused* I thought that was what redundant systems were for... you know, avoid disruptions...

[pphaneuf.livejournal.com]

This is a much higher level. At the level of interrupt handling, Amdahl's Law is restricting us.

[azrhey.livejournal.com]

hey! to sleep with you!

[skjalm.livejournal.com]

I would say that having redundant systems when it comes to (the classic definition of) priority interrupts would most likely result in collisions on the home network ;-)

[azrhey.livejournal.com]

maybe. as I said, it is a domain I am not familiar with, , however, it is my impression that collisions on the home network can, in most cases, be avoided by excellent connectivity and sometimes less traditional ways of building up such network.

but what do I know. I am not even a geek!

[pphaneuf.livejournal.com]

Like she says. That's also one of the reason you mask interrupts while processing one, unless you really know what you're doing (not so difficult, but not something to do mindlessly). You do need a solid interrupt processing strategy, and having an APIC probably wouldn't hurt.

In any case, while I'm handling a priority interrupt, my timer interrupt is masked out, so there's not much multitasking happening. My other task was something along the lines of explaining how recurrent events in an Exchange calendar had to store timezone and daylight savings information in order to work right. That task just got wedged.

[thebabynancy.livejournal.com]

How you ever managed to get laid beyond highschool shocks me daily.


Geek!


*wink*


:) Nancy

[pphaneuf.livejournal.com]

It's funny, 'cause I only gotten laid beyond highschool. In highschool, I wasn't just a geek, I was a nerd. But toward the end, I got better, I even went to my prom and even had fun!

[cpirate.livejournal.com]

In university, the machines we had for our real-time OS class had a little button on the front that would trigger an NMI. They were just two wires that went into a PCI or ISA slot in the machine; nobody knew how they worked, and you could only ever send one per boot, which nobody could ever fix. You certainly couldn't mask them, which made them incredibly handy to get some debugging information.

This also reminds me of the Bastard Operator From Hell stories, where frequent use is made of the "male Non-Maskable Interrupt." You certainly can't ignore that one either :)

[pphaneuf.livejournal.com]

Ohh, the male NMI. Yowser.