Re: a joint letter on low latency and Linux

[David Schleef]

On Thu, Jun 29, 2000 at 09:29:26PM -0400, Paul Barton-Davis wrote:
> >What "hundreds of thousand of lines"?  I think your problem is that
> 
> Lets just start with the ALSA drivers. Currently weighing in at about
> 145,000 lines. All of which *has* to be ported to work with RTLinux.
> 
> Then we can throw in the framebuffer drivers for the video folk, plus
> the various video/tv interfaces, USB for digital audio and MIDI. 
> 
> I think that takes us to at least a couple of hundred thousand lines
> which *must* be ported to RTLinux before any apps, no matter how they
> are designed, can be targeted. Not quite true - most of mine are audio
> only, so just doing ALSA would be fine.

Actually, you might find that the job is easier than you think.
I develop and maintain the data acquisition drivers for Linux,
which are fully compatible with RTAI and RTLinux -- that is,
every feature is available from both hard real-time and user
space, simultaneously.  (Yes, you can control the analog output
from a real-time task while reading analog inputs at 1 M samples/sec
from user space.)

People looking at the source keep asking me: Where's the real-time
support?  When you look at the code, the low level drivers
don't do anything special to work with real-time.  It really
comes down to correctly abstracting your low-level driver
interface, and make the mid-layer handle the details of
who is in charge, Linux or real-time.  In my observation,
making drivers real-time capable has improved the quality of
non-real-time operation.

from the code browsing that I have done, ALSA is architected
fairly decently, and I wouldn't be surprised if the low level
drivers needed very little modification.

dave...