Re: 2.6.24-rt1: timing problems (was [git pull] x86/hrtimer/acpifixes)

[Fernando Lopez-Lezcano]

On Mon, 2008-01-28 at 10:26 -0800, Fernando Lopez-Lezcano wrote:
> On Sun, 2008-01-27 at 05:46 +0100, Mike Galbraith wrote:
> > On Sat, 2008-01-26 at 17:59 -0800, Fernando Lopez-Lezcano wrote:
> > 
> > > Hi Ingo... back to testing. 
> > > History:
> > > 
> > > 2.6.23.x + rt has not been very usable for audio applications. 
> > > 2.6.24-rt1: same so far. 
> > > 
> > > Why: Jack keeps printing "delayed..." messages and has xruns which means
> > > that somehow the timing is delayed more than what jack would think
> > > reasonable. As in the case with an old timing bug, the problem
> > > dissapears when booting the kernel with idle=poll. Other users of Planet
> > > CCRMA are able to replicate the behavior, which goes away with idle=poll
> > > or booting the machine with only one core. As a workaround I have been
> > > packaging 2.6.22.x but now I'm not able to use that as the old rt14
> > > patch, suitably tweaked results in a non working kernel. 
> > > 
> > > So it looks like, again, timing is getting skewed when the jack process
> > > jumps between cpus and thus jack sees timing jumps that are just not
> > > happenning. 
> > > 
> > > This is with a build based on 2.6.24 using as a base the latest Fedora
> > > rawhide source package plus 2.6.24-rt1. 
> > 
> > Do you have a simple testcase?  (one which doesn't entail installing
> > ccrma and becoming an audiophile)
> 
> No, I don't at this point. 
> I'll see if I can cook something simple today... (naively thinking that
> some short C code could test for the clock being actually monotonic
> across cpus). 

Sorry, no luck so far in writing something simple that will fail. I
tried testing for the results from repeated calls to clock_gettime (what
jack uses for timing by default) to actually be monotonic, while a
script uses taskset to force a cpu switch and of course got no errors. 

2.6.24-rt1 with idle=poll works fine, without it I get multiple problems
with the jack internal timing, or least that is what it seems to me from
the symptoms. 

-- Fernando


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/