[linux-audio-user] Re: Realtime Kernel Slows My Clock

[Chuck Martin]

On Wed, Feb 08, 2006 at 11:41:18AM +0100, Florian Schmidt mista.tapas-at-gmx.net |LAU| wrote:

> On Tue, 07 Feb 2006 19:43:18 -0500
> Lee Revell <rlrevell at joe-job.com> wrote:
> 
> > On Tue, 2006-02-07 at 18:46 -0500, Chuck Martin wrote:
> > > On Tue, Feb 07, 2006 at 05:22:23PM -0500, Lee Revell rlrevell-at-joe-job.com |LAU| wrote:
> > > 
> > > > Hmm, IIRC this might have been related to the softirq timer thread being
> > > > preempted - try setting it to the maximum RT priorirty (99).
> > > 
> > > How is that done?  (This is on Slackware 10.0).
> > > 
> > 
> > Not sure, it's in the list archive.
> 
> i think the kernel thread in question is "softirq-timer/0". It is kinda
> resistant against the usual 
> 
> chrt -f -p `pidof "softirq-timer/0"` 

Yes, I tried that, and it didn't work.

> but something like this works:
> 
>  /usr/bin/chrt -f -p 96 `/usr/bin/getpid.sh "softirq-timer/0"`
> 
> where getpid.sh looks like this:
[snip]

I didn't try that with your script, but I did the equivalent by manually
checking the output of "ps aux" for the PID and entering

chrt -f -p 99 3

Here are the first few lines of "ps aux":

USER       PID %CPU %MEM   VSZ  RSS TTY      STAT START   TIME COMMAND
root         1  0.1  0.1   560  244 ?        S    02:19   0:00 init [2]
root         2  0.0  0.0     0    0 ?        S    02:19   0:00 [softirq-high/0]
root         3  0.0  0.0     0    0 ?        S    02:19   0:00 [softirq-timer/0]
root         4  0.0  0.0     0    0 ?        S    02:19   0:00 [softirq-net-tx/]
root         5  0.0  0.0     0    0 ?        S    02:19   0:00 [softirq-net-rx/]
root         6  0.0  0.0     0    0 ?        S    02:19   0:00 [softirq-scsi/0]
root         7  0.0  0.0     0    0 ?        S    02:19   0:00 [softirq-tasklet]
root         8  0.0  0.0     0    0 ?        S<   02:19   0:00 [desched/0]
root         9  0.6  0.0     0    0 ?        S<   02:19   0:00 [events/0]

after trying a PID of 3, I also tried all of the other PIDs from 2 to 9,
as well as every PID for every "[IRQ x]" line in the output of "ps aux",
where x is any number.  None of them fixed the problem.

> I just don't know whether this is a cure for the problem at hand. I
> haven't read the whole thread.

Any other suggestions?  The original problem is that the system clock
is extremely slow, causing the time to be off and things like "sleep 1"
to take 10 to 30 times as long as they should.

Chuck