[prev in list] [next in list] [prev in thread] [next in thread] 

List:       comedi
Subject:    Re: realtime COMEDI
From:       David Schleef <ds () stm ! lbl ! gov>
Date:       2000-10-13 23:03:19
[Download RAW message or body]

On Fri, Oct 13, 2000 at 04:16:10PM -0400, Brent Baccala wrote:
> I think the current implementation is flawed.  A driver that uses
> interrupts will have problems.  The driver won't install an RT-Linux
> interrupt, it'll use a native interrupt.  This means that the RT-Linux
> module will be calling, from a realtime interrupt, a module that uses
> non-realtime interrupts.

Everything is in place for drivers to allocate real-time interrupts.
Some drivers will do this automatically, others not.

> Furthermore, the code as it's written uses a memory buffer (devpriv) to
> get samples back from the realtime process.  As I understand RT-Linux
> (and I'm not an expert) this is a no-no, since the memory buffer might
> get paged out to disk, and we don't do VM paging for realtime processes,
> since this would introduce unacceptable delays.

It is kernel memory allocated with vmalloc().  It doesn't get paged
out.


> I guess the upshot of the whole thing is... first, if COMEDI's going to
> support realtime operation, I suspect the drivers need to be designed
> with this in mind.

They are.  Please read the source.




dave...




[prev in list] [next in list] [prev in thread] [next in thread]