[LAU] Implications of kernel without BKL for RT/Low latency ?

Darrin Thompson darrinth at gmail.com
Tue Nov 2 14:42:02 UTC 2010


On Tue, Nov 2, 2010 at 7:52 AM, Fritz Meissner <meissner.fritz at gmail.com> wrote:
> I noticed this announcement of kernel 2.6.37-rc1
> http://lkml.org/lkml/2010/11/1/82 , and in particular that the BKL is
> largely gone. To my (fairly uninformed) mind, it sounds as though
> taking the lock out of the kernel should make the kernel more
> responsive and maybe make the RT or low-latency kernel easier to do,
> or even unnecessary maybe ? I'd be interested to hear the comments on
> this from our experts.
>

Caveat: I doubt I'm the level of expert you're looking for but I have
written some kernel code and I read lwn too. :-)

I don't think this particular lock has that much impact on audio. The
low-latency gripes with the state of Linux had more to do with the
ability to acquire RT privileges as non root and the risks of allowing
user processes to have so much scheduling power over the rest of the
system.

Also, the lock has been in the process of coming out for years. So
most of the benefits of it being gone have been available to us for
quite some time now. That it's now unused by "core code" is noteworthy
but if your audio driver still uses it then you still have it, and it
might not even matter for your particular use case.

The impression I have is that the latency provided by a default by the
newer kernels is usually quite good. I'm running Arch linux which has
2.6.35.x available and on an i3 box I built this year and things are
going swimmingly for me.

--
Darrin


More information about the Linux-audio-user mailing list