Re: [Agnula-Developers] [ demudi-Feature Requests-723 ] Compile Muse 0.6.3 with givertcap support
Hi,
I let your message appended, and I am trying to respond in a more general
way. Well, we all know that Debian is pretty conservative. It won't be a
problem to get a kernel-patch package included for 2.4 (the way you
describe it, with additions and same interface as 2.6).
But then we still would have to compile patched kernels and get them
through, which might not be that easy.
My main concern is, if it is worth the effort. The module solution for
2.6 is a lot easier to integrate into Debian. Very clean, separate
package, only care has to be taken that it is available for all kernel
flavours.
Then eventually, everyone just wants to use 2.6 in the very near future.
So if there are only small drawbacks in terms of lowlatency behaviour, I
would prefer the 2.6 solution.
Guenter
On 29 Feb 2004, Jack O'Quin wrote:
> One of the permission options provided by this module is `gid=29',
> which grants realtime privileges to any process belonging to group
> `audio'. This fits the Debian structure nicely.
>
> I believe this approach could form *part* of a systematic solution to
> the overall realtime privileges problem. The missing piece is a
> corresponding solution for 2.4 kernels, which will continue being used
> with audio for a long time. It still runs better (when patched) for
> low latencies, though with 2.6.3 latency is improving. AFAIK, 2.4 is
> still the default kernel for sarge, right?
>
> The obvious solution would be to apply the rather large SELinux patch
> from NSA. This back-ports the entire LSM structure to 2.4. I doubt
> the Debian kernel maintainers would approve of that idea, and IMHO it
> is overkill for the small changes needed to grant realtime access.
>
> To my mind, a more practical solution for 2.4 would be something like
> the `schedmlockpatch' kernel patch Kjetil Matheussen posted on LAD a
> while back.
>
> For this approach to be adopted by a major distribution like Debian
> multimedia, I would recommend doing some additional work to make the
> permission methods for both kernel versions look and behave similarly.
> I would control them through the same interface in /proc/sys/kernel
> (or some place like that).
>
> I have done some thinking about this. There was a good discussion on
> LAD a few months back. If there is interest I will make a specific
> proposal, and maybe come up with a modified 2.4 patch for people to
> experiement with.
> --
> joq
>
Reply to: