On Mon, 15 Jan 2001, Andrew Morton wrote:
> How serious an issue is the latency introduced by fbdevs?
>
> I've taken a look at what's going on in there and it seems
> that it's fairly straightforward to fix (*). It definitely
> _needs_ fixing - it's putrid.
>
> The one drawback of the approach I'm thinking about is that
> kernel printk's from interrupt context wouldn't appear
> on the console (but they would make it to the logfile). Or
> maybe they will, but delayed a bit.
>
> Question is: is this worth the effort, or are people
> OK with just running X or vgacon all the time?
If latencies can be fixed, it would be nice.
Perhaps someone would want to use a Linux box in fbdev mode as a realtime
video processing box. (30-60fps without frame losses).
With X it is currently not easy (or impossible ?) to achieve zero frame losses,
due to the X process running as a normal user process.
Perhaps this can be achieved using DRI, but I'm not an expert in that field.
BTW: Andrew keep up with your great LL stuff !
Haven't had the time to test it, but has anyone done some latencytest
graphs on the latest 2.4-LL kernels ?
If yes, please post the URL here.
cheers,
Benno.
>
> Thanks.
>
>
> (*) Easy to fix for Voodoo 3, TNT2 and G200. That's
> all I have :)
--
http://www.linuxaudiodev.com The Home of Linux Audio Development
